This part made me realize why tuning the performance of the modern systems is highly reliant on the hardware awareness.

The comparison of multicore and NUMA architecture explains the direct influence of the evolution of hardware on the design of OS. It is obvious that operating systems can no longer settle on homogeneous memory access, which complicates scheduling and management of memory.

Interrupts were one of the roles that I found interesting since the CPU is allowed to work effectively rather than having to check the devices constantly. This is a sort of early version of event-driven design that is heavily utilized in contemporary software systems.

The user considers the OS to be concerned with convenience and ease of use, whereas the system is concerned with efficiency and resources in use. Such a difference can justify why certain OS choices are frustrating to users and are needed to have a stable system.

This is Ananth Kumar Madala and I got access to the text book and this is my annotation tool

The annotation is done to ensure that I am able to get into the textbook and that the annotation tool is functional. I can open the book, go to the cover page and do some annotations without any complications. This proves that I am in full access of textbook to complete future reading and annotation work.

The operating system as a very crucial layer that coordinates the hardware resources and allows the application programs to operate effectively. It also dwells on the necessity of knowledge of the computer hardware architecture such as the CPU, the memory, the storage and the I/O devices to comprehend the role of an operating system. It states that operating systems are built in small parts to accommodate complexity, which makes them understandable and resourceful. It is based on this structure that the study of system design, data structures, and open-source operating systems are built.

JSONSchemaBench: Structured Output Benchmark for Language Models

Core Contribution

• Benchmark introduction: JSONSchemaBench comprising 10K real-world JSON schemas for evaluating constrained decoding frameworks

  "We introduce JSONSchemaBench, a benchmark for constrained decoding comprising 10K real-world JSON schemas that encompass a wide range of constraints with varying complexity"

• Three-dimensional evaluation framework assessing efficiency, coverage, and quality of constrained decoding approaches

  "We present an evaluation framework to assess constrained decoding approaches across three critical dimensions: efficiency in generating constraint-compliant outputs, coverage of diverse constraint types, and quality of the generated outputs"

• Six frameworks evaluated: Guidance, Outlines, Llamacpp, XGrammar, OpenAI, and Gemini

  "We evaluate six state-of-the-art constrained decoding frameworks, including Guidance, Outlines, Llamacpp, XGrammar, OpenAI, and Gemini"

Key Findings

Efficiency Results

• 50% speedup potential from constrained decoding over unconstrained generation

  "Constrained decoding can speed up the generation process by 50% compared to unconstrained decoding"

• Guidance achieves best throughput through guidance acceleration technique

  "Guidance achieves even higher efficiency, which it accomplishes by fast-forwarding certain generation steps with its guidance acceleration"

• Compilation time varies significantly: Outlines has "significantly higher compilation time" compared to Guidance and Llamacpp which have "minimal grammar compilation time"

Coverage Results

• Coverage disparity: Best framework supports twice as many schemas as worst

  "Frameworks demonstrate significant differences in their actual support for real-world JSON schemas, with the best framework supporting twice as many schemas as the worst"

• Guidance leads empirical coverage on 6 out of 8 datasets

  "Guidance shows the highest empirical coverage on six out of the eight datasets"

• Compliance rate critical: Guidance demonstrates "highest compliance rate across all datasets, making it the most reliable option for ensuring schema compliance"

Quality Results

• Performance improvement: Constrained decoding improves downstream task accuracy up to 4%

  "Constrained decoding consistently improves the performance of downstream tasks up to 4%, even for tasks with minimal structure like GSM8k"

• Guidance best for quality: "Guidance consistently delivers the best performance across all tasks, with approximately a 3% improvement over the LM-only approach in every task"

Core Concepts

Coverage Definitions

• Declared Coverage: Schema processed without rejection or runtime errors

  "A schema is considered declared covered if the framework processes the schema without explicitly rejecting it or encountering runtime errors such as exceptions or crashes"

• Empirical Coverage: Experiments show framework produces schema-compliant outputs

  "A schema is considered empirically covered if our experiments show that the constraints generated by the framework result in LM outputs that are schema-compliant"

• True Coverage: Constraints precisely equivalent to JSON Schema definition

  "A schema is considered truly covered if the framework produces constraints that are precisely equivalent to the original JSON Schema definition"

• Compliance Rate: Ratio of empirical to declared coverage

  "Compliance Rate = C_Empirical/C_Declared... estimates the reliability of the constrained decoding framework in guaranteeing compliance given it accepts a given schema"

Failure Modes

• Over-constrained: Framework rejects valid JSON instances

  "A framework is over-constrained if it rejects JSON instances that are valid according to a given JSON Schema. This means the engine is too strict and excludes outputs that should be allowed"

• Under-constrained: Framework allows invalid JSON instances

  "A framework is under-constrained if it allows JSON instances that are invalid according to a given JSON Schema. This means the engine is overly permissive and allows outputs that should be rejected"

Evaluation Methodology

Efficiency Metrics

• Grammar Compilation Time (GCT): Time spent on grammar compilation
• Time to First Token (TTFT): Time from generation start to first token
• Time per Output Token (TPOT): Average time per token after first

Dataset Composition

• 10 dataset suites with varying complexity:

  "We split the data into five collections based on the schema size: trivial, small, medium, large, ultra"

• Sources include:

• GitHub schemas (6,000 from Baazizi et al., 2021)
• JSON Schema Test Suite (official test cases)
• Schema Store (largest independent collection)
• GlaiveAI function calling dataset (2,000 schemas)
• Kubernetes configuration files
• Washington Post ANS specification

Experimental Setup

• Model used: Llama-3.1-8B-Instruct for efficiency, Llama-3.2-1B-Instruct for coverage
• Hardware: Single NVIDIA A100-SXM4-80GB GPU with AMD EPYC 7543 CPU
• Validation: jsonschema Python library with Draft2020-12, format checks enabled
• Generation: Greedy decoding, zero temperature, single run, 40s timeouts

JSON Schema Test Suite Analysis

• 45 categories covering JSON Schema features (43 after filtering)

  "The test suite organizes its test cases into 45 categories, each of which corresponds to a feature of JSON Schema"

• Guidance dominates:

  "Guidance outperforms other engines at all coverage levels, achieving full coverage on 13 categories and moderate coverage on 21"

• Failure patterns vary:

  "Outlines, Llamacpp, and Guidance follow a consistent failure pattern, with most errors occurring during compilation and over-constrained failures being more frequent than under-constrained ones. In contrast, XGrammar minimizes compilation errors but shows the highest number of under-constrained failures"

Quality Evaluation Tasks

Three Reasoning Tasks

• Last Letter: CoT reasoning + answer in a-z

  "Input: Ian Peter Bernard Stephen Output: nrdn"

• Shuffle Objects: CoT reasoning + answer in A-E

  "Input: Sequence of exchanges among individuals + choices Output: A-E"

• GSM8K: CoT reasoning + answer as integer/float

  "Input: Basic calculation problems Output: Number, e.g. 8"

• JSON structure: All tasks use {"reasoning": <reasoning>, "answer": <final answer>} format

Problem Context

Motivation

• Machine-oriented applications require structured outputs

  "Unlike traditional natural language processing (NLP) tasks where the output is aimed at review by humans, output in these applications is often consumed by machines such as controller and service APIs"

• Probabilistic nature problematic:

  "However, the LM generation process is probabilistic and does not provide guarantees on the output's structure, making it challenging to deploy LMs in applications requiring structured inputs and high reliability"

• JSON Schema as standard:

  "JSON Schema has emerged as a key specification language for constrained decoding... Commercial LM providers, such as OpenAI, have embraced constrained decoding by incorporating support for JSON Schema directly into their APIs"

Research Gap

• Evaluation under-explored:

  "The evaluation of constrained decoding remains an under-explored topic, with no consensus on what defines the effectiveness of constrained decoding"

• No framework comparisons: Prior studies "fail to provide comparisons across different constrained decoding frameworks"

• Limited benchmarks: Previous benchmarks "narrowly focused on specific tasks or rely on formal-grammar–based artificial setups, that have unclear relevance to real-world use cases"

Technical Implementation

Constrained Decoding Algorithm

• Core mechanism: Token masking at each generation step

  "Constrained decoding intervenes in the decoding process of LMs by masking out invalid tokens based on given constraints and prefix tokens"

• Algorithm steps: Update constraint state → compute mask → calculate logits → apply mask → sample token → append to output

Addressing Coverage Bias

• Intersection-based metrics: Calculate efficiency only on schemas all engines support

  "To ensure fairness, we calculate efficiency metrics on the intersection of covered instances across all engines"

• Rationale: "Engines with lower coverage often process simpler, shorter schemas, which naturally compile and generate faster"

Key Tools & Frameworks

Open Source Engines

• Guidance (Guidance AI, 2023): Dynamic constraint computation, token healing
• Outlines (Willard & Louf, 2023): Regex-based constraints
• Llamacpp (Gerganov & al., 2023): Grammar module, GGML BNF
• XGrammar (Dong et al., 2024): Concurrent compilation with pre-filling

Closed Source APIs

• OpenAI: JSON Schema support in API
• Gemini: Limited schema support (<1% on some datasets)

Validation & Testing

• jsonschema Python library (Berman, 2025): Draft2020-12 validation
• JSON Schema Test Suite (JSON Schema Org, 2024): Official correctness tests
• Bowtie (2025): Cross-implementation comparison tool

Related Work

JSON Schema Collections

• Baazizi et al., 2021: 6,000+ schemas from GitHub repositories
• Attouche et al., 2022: Witness generation algorithm evaluation
• Schema Store: Largest independent JSON schema collection

Constrained Decoding Research

• Early methods:
• Deutsch et al., 2019: General-purpose constrained sequential inference
• Shin et al., 2021: Constrained language models as semantic parsers

• Scholak et al., 2021: PICARD for incremental parsing

• Optimization frameworks:

• Beurer-Kellner et al., 2023: Prompting as programming query language
• Kuchnik et al., 2023: ReLM validation
• Zheng et al., 2024: SGLang structured execution

Quality Concerns

• Geng et al., 2023: Grammar-constrained decoding without finetuning
• Tam et al., 2024: Study on format restriction impacts
• Kurt, 2024: Response to performance decline concerns, argues issues stem from "inadequate prompting, insufficient contextual information, and poorly crafted schemas"
• Geng et al., 2024: Tokenization ambiguity issues

Applications

• Yao et al., 2023: Web navigation (ReAct)
• Polak & Morgan, 2024: Materials data extraction
• Schick et al., 2023: Toolformer for tool use

Advanced Features

Optimization Techniques

• Grammar caching: Reuse compiled constraints
• Parallel execution: "Mask computation can run in parallel with the LM's forward pass, and grammar compilation can be performed concurrently with pre-filling computations"
• Speculative decoding: Constraint-based approach (GuidanceAI, 2024)
• Token healing: Guidance's technique for handling tokenization boundaries (GuidanceAI, 2024)

JSON Schema Features

• Most common keywords: type, properties, required, description, items, enum
• Format constraints: date-time, email, uri, uuid (each complex to implement)
• Advanced features: $defs, $ref, if-then-else, minItems, maxItems, pattern matching

Limitations & Future Work

Current Challenges

• Under-constraining tradeoffs:

  "Under-constraining effectively delegates responsibility to the LM, which may produce valid output despite a lack of strict constraints... requires careful implementation and transparency to ensure reliability"

• Feature representation: "Not all features are equally represented in real-world schemas... strong or weak performance on specific features can have disproportionate impacts depending on their prevalence"

• Test suite limitations: "No straightforward correspondence between test suite performance and empirical coverage"

Aspirational Goals

• GitHub-Ultra dataset: Retained "as an aspirational target for future advancements" despite being too hard for current frameworks

• Format validation: Paper excluded 'format' tests but notes "We hope to extend this work to include these optional tests in a follow-up"

• Formal verification: True coverage requires "formal verification method that is capable of exhaustively comparing the schema's semantics against the framework's implementation"

Repository & Resources

• Benchmark release: https://github.com/guidance-ai/jsonschemabench
• JSON Schema specification: Draft 2020-12 (Wright et al., 2022)
• Model used: Llama-3.1-8B-Instruct (Grattafiori et al., 2024)

It's interesting to me that he uses the logic of the Earth being round to know where to travel and understand the geography of the world. For this time period, I think it's impressive how he is using this scientific reasoning.

https://www.americanyawp.com/reader/

The structures of the paragraph are very clean as both important texts are underlined and all paragraphs broken up for easier readability for everyone with a large font. This especially helps users with reduced cognitive load and lets users read the page any way they want.

The advertisement here, while not really on the reader's path, may throw off some users and disrupt their reading flow, which can undermine accessibility specially with a small text.

A large heading at the beginning of the review blog which helps users understand the topic quickly and navigate which goes by the rule of Perceivable understandable principle making the website seem transparent.

A very well-integrated Navigation link at the top of the article that lets users move between different parts of the page. This successfully obeys the operable principle as the interface helps navigate through the website more easily and is much more accessible to anyone reading the blog.

The image here shows a snapshot of the vehicle, with a clear price description and even a link to press to take the user to the full price and specs of this vehicle, making the website highly accessible for users and easy to navigate. While the description of the vehicle might seem confusing, it is due to the long name of the car itself rather than a website issue, which also proves that Top Gear review makes their images descriptive to enhance visual understanding.

The Main heading helps readers easily understand the entry point of this review, so they don't have to wonder what they should expect to read. This helps with the receivable and understandable principle as it helps users get an understanding of the topic and navigation.

When writing about culture making sure you have some awareness of the culture is very important. Understanding what your writing about helps you present your ideas and have them sound more confident. Its great to make sure your treating people fairly talking about any culture. Writing is more credible when ethics are taken seriously.

"Plus grades will be assigned if a higher benchmark for the next grade has been earned in one or more other categories." So how exactly is our final grade going to be calculated?

Are there any other guidelines for the eye protection, does it have to be goggles with an elastic band?

"without energy provided by the cell"

If a ethnographic writing is never really objective, how can a writer include their own viewpoint while still representing the culture correctly? This is significant because it affects how clear and believable the essay is

They are mostly empty as they are set in TypeMeta and that are not filled in memory unless someone/decoder explicitly fills it.

couldn't understand fully, but ok for now.

get some example?

So basically think of the GVK as API contract and the Go Type as Implementation.

So, basically it should be only one for the implementation so that it can decode without being ambiguous but then again to maintain multiple versions, it has to support multiple types.

① Five Leads (Beş Elektrot) I, II, III, aVR, aVL ve aVF derivasyonlarını içerir.

② Fifth electrode (Beşinci elektrot) V1 ile V6 arasına yerleştirilebilir.

③ V1 kullanımı Aritmileri tespit etmek için kullanılır.

④ V3–V5 kullanımı İskemi tespiti için kullanılır.

Monitorizasyon, klinik kararların doğruluğunu ve hassasiyetini artırır.

① Autosomal dominant inheritance Otozomal dominant kalıtım

② It is an acute hypermetabolic state that occurs in muscle tissue, rare Kas dokusunda ortaya çıkan nadir bir akut hipermetabolik durumdur

③ Recent studies mention mutations in the skeletal muscle Ca channel and Ryanodine receptor genes on chromosome 19 in humans Son çalışmalar, insanlarda 19. kromozomda bulunan iskelet kası Ca kanalı ve Ryanodin reseptör genlerindeki mutasyonları bildirmektedir

④ May develop with the onset of anesthesia or in the postoperative period Anestezi başlangıcında veya postoperatif dönemde gelişebilir

Postoperatif hipoksinin başlıca nedenleri; hipoventilasyon, ventilasyon-perfüzyon uyumsuzluğu ve artmış pulmoner şantlardır

① In the early postoperative period; hypoxia, hypercapnia and aspiration are observed most frequently Erken postoperatif dönemde en sık hipoksi, hiperkapni ve aspirasyon görülür.

① Choosing the anesthesia plan to be followed according to the risk factors determined in the medical history and the choice of the patient, making perioperative and postoperative care more effective and cheaper Tıbbi öyküde belirlenen risk faktörlerine ve hastanın tercihine göre uygulanacak anestezi planının seçilmesi, böylece perioperatif ve postoperatif bakımın daha etkili ve daha ekonomik hale getirilmesi

② Reducing the mortality and morbidity of surgery Cerrahinin mortalite ve morbiditesinin azaltılması

③ To enable the patient to return to normal life as early as possible Hastanın normal yaşamına mümkün olan en kısa sürede dönebilmesinin sağlanması

① Obtaining information about the patient's medical history, physical and mental status and determining the necessary tests and consultations Hastanın tıbbi öyküsü, fiziksel ve ruhsal durumu hakkında bilgi edinilmesi ve gerekli tetkiklerin ve konsültasyonların belirlenmesi

② Choosing the anesthesia plan to be followed according to the risk factors determined in the medical history and the patient's choice Tıbbi öyküde belirlenen risk faktörlerine ve hastanın tercihine göre uygulanacak anestezi planının seçilmesi

③ Giving detailed information to the patient and obtaining consent Hastaya detaylı bilgi verilmesi ve aydınlatılmış onamın alınması

④ Educating the patient about anesthesia, preoperative care and pain management to reduce anxiety and facilitate recovery Kaygıyı azaltmak ve iyileşmeyi kolaylaştırmak amacıyla hastanın anestezi, preoperatif bakım ve ağrı yönetimi konusunda eğitilmesi

It's honestly a bit unsettling how quickly he reduces these people to potential servants instead of seeing them as human beings who deserve equal rights. As in, his first instinct was instantly control over these people to use them for his own benefits and beliefs. I can see now why disagreements arise in the future; there was lack of mutual respect from the very start.

Hollister has an inaccessible navigation menu, as there is no option to make text larger. However, if there is an option to do so, it is difficult to find. The annotated text is important information regarding the parameters of Hollister's sale dates. Therefore, there should be an option to make the text larger. This is a poor practice of web accessibility.

There is no option to space out the text, or the color/design bubbles. This may prevent users with mobility challenges from effectively clicking on the desired item. This is a poor practice of web accessibility.

Below are styles titled Our Most Wanted Looks. To see the styles, you must click the arrow on the right hand side. There is no "slide" function allowing users to see the hidden styles without clicking on the arrow. This may be a hindrance to users with mobility challenges, and as a result, this is a poor practice of web accessibility.

I have turned on the "high contrast" feature on this webpage, which may help visually impaired individuals view text clearly. This is a good practice of web accessibility.

What's surprising is how it seems like they lacked a sense of fear when it came to the thought of something unfamiliar to them having the possibility of being poisonous. I guess this shows how differently people in the past interacted with nature compared to how we interact with it today, but I guess it is also because of modern common knowledge that has developed.

It's frustrating to know that colonizers always took credit for things that natives worked hard to achieve and then proceed to erase the people who were responsible for creating and caring for this land in the first place. Let us not forget who this land had belonged to and who we need to keep in mind when we appreciate how beautiful of a land we have.

I do work a full-time job, so this semester I will prioritize time management and do my best to find time during the day to read for my assignments. Taking the time to plan before I read will help me breakdown my readings better.

I wonder if any of these original stone and mud buildings still exist today or if the ones we can see today are reconstructions of Spaniard settlements.

I wonder if people ever ended up questioning these rituals, then again I understand that this wasn't just any tradition but one deeply rooted in the belief of it being necessary for survival. And I don't think they could have ever reached that point of being able to evolve their beliefs because of colonization.

It's cool that chinampas were invented so long ago and yet are still being used today. I got to see one in real life, and it almost feels fairytale with how much beautiful greenery there is. It's incredible that they found this as a sustainable way to feed their population because it shows just how advanced their engineering was back then.

Black theology does have issues though especially with hiw it ignores black women and black sexuality in general

Cone also criticized Black churches for trying to strike a reconcilitory tone that tried to reach white respectability

pushed the fact that Jesus was black as Jesus only stands on the side of the oppressor

James Cone pushed the white churches were proponents of oppression as they either pushed white supremacist dogma or igorned the suffering of Blacks

Black churches were important to African Americans because they were centers for political activism especially the civil rights movement

with this the 19th century saw Blacks establish their own Christian churches and denominations which started first with Richard Allen in 1816.

so instead colonists made laws and teachings that justified slavery while also converting the slaves to it

History of the Black Church was initially poised by reluctance by slavers to convert their slaves as they believed that the British law mandating freedom for all Christians would mean freedom for them

The black church started as slavers converted many African Americans to Christianity through force to justify slavery

This section of text answers the question "Are there any concerns presented about the results?" The section of limitations take into account that the primary results are sourced from self reports and reports of peers. The article understands that the accuracy of this study can be questioned due to the nature of not being able to directly check the accuracy of these reports but assures that they sourced detailed responses and constistancy within those responses.

From the conditions of frontier life came intellectual traits of profound importance. The works of travelers along each frontier from colonial days onward describe certain common traits, and these traits have, while softening down, still persisted as survivals in the place of their origin, even when a higher social organization succeeded. The result is that, to the frontier, the American intellect owes its striking characteristics. That coarseness and strength combined with acuteness and inquisitiveness, that practical, inventive turn of mind, quick to find expedients, that masterful grasp of material things, lacking in the artistic but powerful to effect great ends, that restless, nervous energy, that dominant individualism, working for good and for evil, and withal that buoyancy and exuberance which comes with freedom -- these are traits of the frontier, or traits called out elsewhere because of the existence of the frontier.

As the eastern lands were taken up, migration flowed across them to the west. Daniel Boone, the great backwoodsman, who combined the occupations of hunter, trader, cattle-raiser, farmer, and surveyor -- learning, probably from the traders, of the fertility of the lands of the upper Yadkin, where the traders were wont to rest as they took their way to the Indians, left his Pennsylvania home with his father, and passed down the Great Valley road to that stream.

Learning from a trader whose posts were on the Red River in Kentucky of its game and rich pastures, he pioneered the way for the farmers to that region. Thence he passed to the frontier of Missouri, where his settlement was long a landmark on the frontier. Here again he helped to open the way for civilization, finding salt licks, and trails, and land.

The effect of the Indian frontier as a consolidating agent in our history is important. From the close of the seventeenth century various intercolonial congresses have been called to treat with Indians and establish common measures of defense. Particularism was strongest in colonies with no Indian frontier. This frontier stretched along the western border like a cord of union. The Indian was a common danger, demanding united action. Most celebrated of these conferences was the Albany congress of 1754, called to treat with the Six Nations, and to consider plans of union. Even a cursory reading of the plan proposed by the congress reveals the importance of the frontier. The powers of the general council and the officers were, chiefly, the determination of peace and war with the Indians, the regulation of Indian trade, the purchase of Indian lands, and the creation and government of new settlements as a security against the Indians.

Long before the pioneer farmer appeared on the scene, primitive Indian life had passed away. The farmers met Indians armed with guns. The trading frontier, while steadily undermining Indian power by making the tribes ultimately dependent on the whites, yet, through its sale of guns, gave to the Indian increased power of resistance to the farming frontier. French colonization was dominated by its trading frontier; English colonization by its farming frontier. There was an antagonism between the two frontiers as between the two nations. Said Duquesne to the Iroquois,

Little by little he transforms the wilderness, but the outcome is not the old Europe, not simply the development of Germanic germs, anymore than the first phenomenon was a case of reversion to the Germanic mark. The fact is, that here is a new product that is American.

Now, the frontier is the line of most rapid and effective Americanization. The wilderness masters the colonist. It finds him a European in dress, industries, tools, modes of travel, and thought. It takes him from the railroad car and puts him in the birch canoe. It strips off the garments of civilization and arrays him in the hunting shirt and the moccasin. It puts him in the log cabin of the Cherokee and Iroquois and runs an Indian palisade around him. Before long he has gone to planting Indian corn and plowing with a sharp stick; he shouts the war cry and takes the scalp in orthodox Indian fashion. In short, at the frontier the environment is at first too strong for the man. He must accept the conditions which it furnishes or perish, and so he fits himself into the Indian clearings and follows the Indian trails.

Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

Learn more at Review Commons

Reply to the reviewers

Reviewer #1 (Evidence, reproducibility and clarity (Required)): The key conclusions are solid. All the claims are supported by quality data. The content is rich, and no additional experiment is needed. The data and methods are properly presented for reproduction. The experiments are adequately replicated. One comment on statistical analysis is listed below.* *

__Summary:_ ___ This manuscript investigates how Drosophila immune pathways contribute to defense against a range of filamentous fungi with distinct ecological strategies. The work provides novel insights into Toll pathway activation through pattern recognition receptors and danger signals, relative roles of melanization, phagocytosis, and effects of antimicrobial peptides, and particularly the immune evasion strategy of E. muscae via protoplast formation. These findings are of broad relevance to insect immunology, host-pathogen interactions, and evolutionary biology. * The study is well designed, the experiments are carefully executed, and the manuscript is clearly written. It is novel to demonstrate that E. muscae evades immune recognition via protoplast formation. However, some aspects of clarity and discussion of limitations could be improved before publication.** *

We thank the reviewer of the positive assessment of our manuscript.We thank the reviewer of the positive assessment of our manuscript.

Major comments: 1) The Abstract is informative but a bit too long. Consider condensing some sentences and highlighting the novel contributions (e.g., role of protoplasts in immune evasion.).* *

Good points. We have reduced the abstract. The sentence is 'Our study also reveals that the fly-specific obligate fungus Entomophthora muscae employs a vegetative development strategy, protoplasts, to hide from the host immune response.'

We believe that the role of protoplasts is already mentioned in the abstract.

2) The Results may use more mechanistic links. For instance, the section on E. muscae immune evasion could more explicitly connect the morphological findings (protoplasts, lack of cell wall) with specific immune recognition failures.* *

Our article is a comparison of Drosophila host defense against fungi with various life styles. This obviously complexify the presentation of the results. We have made the maximum of effort to explain our data with clarity. We believe that having two successive sections entitled 'Natural infection with E. muscae barely induces the Toll pathway' followed by ' __Entomophthora muscae hides from the host immune response using a vegetative development strategy'____ __expose well the idea that E. muscae has a specific hiding strategy. We did not change this part.

3) Please clarify statistical analyses used for survival data (e.g., log-rank tests, multiple testing corrections). * We have clarified the statistical analysis in the method part. The sentence is 'Statistical significance of survival data was calculated with a log-rank test (Mantel-Cox test) comparing each genotype to w*1118 flies'.

__Minor comments:____ __ Abstract: 1) "The infection outcome depends on the complex interplay between insect immune defenses and fungal adaptive strategies." could be simplified to: "Infection outcomes depend on the interplay between insect immunity and fungal adaptation." 2) Replace "our study uncovers" with "we show" for more concise phrasing. Reduce phrases like "our study reveals" or 'we conclude" in other parts of the manuscript. * Results: p. 5: phrase "survival upon natural infection... reveals the major contribution" → reword to avoid passive tone. p. 10: clarify "vesicles push the membrane outwards" with more precise terminology (e.g., budding, extrusion). * Discussion: p. 20: streamline sentence beginning "These observations provide a mechanistic basis..." (currently too dense).

We have taken in consideration all these comments. Note that we removed in the revised version the sentence "The infection outcome depends on the complex interplay between insect immune defenses and fungal adaptive strategies." To shorten the abstract, we have removed the sentence 'These observations provide a mechanistic basis for future exploration.'

**Referee cross-commenting*** *

I agree with the comments of the other two reviewers.* *

__Reviewer #1 (Significance (Required)):____ __

This manuscript investigates how Drosophila immune pathways contribute to defense against a range of filamentous fungi with distinct ecological strategies (generalists, specialists, opportunists). By leveraging a comprehensive panel of genetically defined fly lines and standardized infections, the authors provide a demonstration that the Toll pathway is the predominant systemic antifungal defense, extending classical findings into a comparative framework across fungal lifestyles. The work provides novel insights into Toll pathway activation through GNBP3 and fungal proteases sensed by Psh, while also dissecting the relative contributions of melanization, phagocytosis, and antimicrobial peptides to host protection. Of particular note is the compelling demonstration that the fly specialist E. muscae can evade immune recognition through protoplast-like vegetative forms, minimizing cell-wall exposure and thereby escaping Toll activation.* *

My expertise and limitations: * Insect biochemistry and molecular biology, with particular focus on innate immunity, serine protease cascades, melanization, and host-pathogen interactions. I also have experience with genetic, biochemical, and functional approaches to dissecting immune signaling pathways in model insects. However, I do not have sufficient expertise to critically evaluate advanced statistical analyses.** *

__Reviewer #2 (Evidence, reproducibility and clarity (Required)):____ __

In this work the authors describe the contribution of distinct immune responses in Drosophila melanogaster to systemic and natural infections with 5 fungal species with different lifestyles some being generalists infecting a broad range of insects while others being more specialists or opportunistic. The authors used several well characterized Drosophila mutants of the Toll, Imd, phagocytosis and melanization responses to address this question. They show that Toll pathway is the key player in anti-fungal resistance in both natural and septic infections, whereas melanization plays a minor role mainly during natural infections possibly to limit fungal invasion through the cuticle. The authors show elegantly using different combinations of mutants for antimicrobial peptides genes with antifungal activities that Bomanins and Daisho (1 and 2) are the main Toll effectors mediating resistance to fungi but the authors did not find specific fungus-by-gene interaction, but rather antifungal peptides seem to act in a more general fashion against the fungi tested with significant redundancies between certain classes. Interestingly the authors show that while generalists like Beauveria and Metarhizium strongly activate the Toll pathway, the specialist E. muscae weakly activates the pathway and the opportunistic A. fumigatus does not activate the pathway, indicating that certain fungal species are able to evade sensing by immune pathways. In the context of the Toll activation, the sensor protease Psh and not GNBP3 seem to be the main trigger of the pathway.* *

__Minor comments____ __ This is an interesting work that compares the contributions of different arms of the fly immune response to 5 fungal species with diverse lifestyles. The use of different lines with different combinations of mutant genes is a strength to highlight the relative contribution of each immune response. Some of the data obtained is intriguing and warrants more future investigations such as the distinct phenotypes of ModSp and GNBP3 mutants in E. muscae infections. The methodology is robust and the conclusions are supported with good experimental evidence. I do not see any major concerns with the work. I just have some minor comments listed below* *

We thank the reviewer for the positive comments on our manuscript. 1- Statistical significance should be indicated on Figures 1 and 2, although it appears in the legend.

We have added statistical significance on Figures 1 and 2.

2- It is not very accurate to use the term resistance of the different mutants to infections with the diverse fungal species in Figures 1 and 2 especially that the authors have reported only survival data in these figures and have not measured fungal proliferation in infected flies (although they did that in later figures). It is more accurate to mention that the mutants flies have different levels of tolerance rather than resistance to fungal infections.* *

We agree that we cannot use the term 'resistance' in Figures 1 and 2, since this term has now a more restricted meaning in the community. We have replaced the term 'resistance' by 'host defense' or 'surviving' through the text to avoid the confusion, except when the bacterial load was monitored.

3- The authors show that Toll is over-activated in PPO1/PPO2 double mutant possibly through a negative feedback mechanism. However, there could be another explanation for this observation: For instance, the increased fungal proliferation in the PPO double mutant results in increased protease secretion by fungi enhancing Psh activation! Also, how can fungi manage to proliferate in this double mutant if Toll is overactivated? Could it be that Toll overactivation is triggering a fitness cost?* *

The reviewer raises a good point. It is difficult to reconcile the susceptibility of PPO1/2 mutants to fungi taking in consideration the higher Toll activation. The higher activation of Toll could be deleterious and We clearly observed higher Toll pathway activation in PPO1/2 flies upon clean injury (Fig. S9C) or injection of dead spores (data not shown). Thus, this higher expression cannot be only explained as a consequence of higher fungal growth.

4- In Lines 654-655, it is not accurate to say that E. muscae protoplasts are not detected by the immune response since E. muscae natural infections triggers Drs expression at 24 hpi and there is possibly some melanization taking place since PPO1 and PPO2 are required for defense against this fungus. A more accurate explanation is that this fungus is possibly more resistant to the effectors of the host immune response than the other fungi. I think a major point that the authors might have missed to consider in the discussion of their data is that the different fungi used herein may exhibit different levels of resilience to the effector reactions of the host such as AMPs and melanin deposition* *

*The observation that injection of E. muscae protoplasts do not trigger an immune response above the level of clean injury is a strong argument that support our view that E. muscae protoplasts are not immunogenic. The reviewer is correct by underlying the small but significant induction of Drs at 24h post natural infection. We hypothesize that this could be due to mechanical injury associated with the entry of E. muscae. We have added a sentence to underline the possibility raised by the reviewer: 'Although we cannot rule out that the high pathogenicity of E. muscae may be partly due to the fungus's increased resilience, we favor the interpretation that it is instead mainly driven by its capacity to evade immune detection.'

__Reviewer #2 (Significance (Required)):____ __

Although the importance of Toll pathway and melanization in antifungal immunity is not new per se, this work adds to this knowledge by showing that Toll has the upper hand in anti-fungal immunity and that the strength of Toll pathway activation and its effector capacity may vary depending on the type of invading fungus. The work also highlights that certain fungi may employ a delayed switch to hyphal growth to reduce the presence of cell wall sugars as a mechanism to evade immune recognition. Overall, this work significantly adds to the knowledge of Drosophila immunity and raises some interesting questions related to the evolution of host-pathogen interactions and to the complex functions of serine protease cascades regulating Toll and melanization. This work will be of interest to a broad audience in the field of host-pathogen interactions *

__Reviewer #3 (Evidence, reproducibility and clarity (Required)):____ __

This is a clearly written manuscript on the immune effector mechanisms regulating Drosophila melanogaster host defense against a broad range of fungal pathogens, including entomopathogenic and saprophytic filamentous fungi. The authors systematically dissect the contribution of major arms of Drosophila immunity, including cellular and humoral responses and melanization and potential mechanisms of cross talk using genetic tools and reporter lines. They also go into detail to characterize the contribution of upstream activators of these responses by fungal PAMPs and the role of antimicrobial effectors (AMPs) in fly susceptibility. * They conclude for no important role of phagocytosis in host defense. Instead, they find important contributions of Toll pathway mainly through the detection of fungal proteases by Persephone rather than b-glucan detection by GNBP3. They also demonstrate that Toll activation is proportional to the virulence of the fungal pathogen, showing little activation of this response by Aspergillus fumigatus. Finally, they identify melanization as another line of host defense that restricts pathogen dissemination and protects fly from invasive fungal disease. A very interesting part of this study is the identification of a virulence strategy of the obligate fungus Entomophthora muscae, which employs a vegetative development strategy, by making protoplast that avoid immune recognition by masking immunostimulatory cell wall molecules to avoid immune recognition by Toll pathway until the very last stage of invasive growth. Overall, this is a very interesting study on host-pathogen interplay in Drosophila, shedding light onto novel pathogenetic mechanism employed by entomopathogenic fungi to adapt to their hosts.** *

We thank the reviewer for his positive assessment.

__Major comments for the authors:____ __ 1. The use of reporter fungal strains to capture the dynamic interplay of the pathogen and the different arms of the immune system precludes firm conclusions on the contribution of various immune response to infection. This should be emphasized in the discussion* *

Unfortunately, we did not fully understand this point. Note that we monitored both survival and when possible fungal load (B. Beauveria, E. muscae and M. anisopliae for Toll; and B. Beauveria, and M. anisopliae for melanization) allowing to state that Toll and Melanization are contributing to host defense by limiting fungal growth.

2. The route of infection and the method employed to inject fungal spores has an impact on the effector pathways being activated. For example, pricking introduces spores less efficiently in the hemolymph compared to microinjection. The inoculum size in case of microinjection also has profound impact in understanding the role of cellular and humoral immunity during the infection course. For example, the lack of Toll activation in the natural infection with A. fumigatus does not mean that this pathway is not important in host defense against this pathogen.

We fully agree and expected to clarify this different outcome between septic injury and natural infection. In the case of A. fumigatus, we confirm that Toll is important upon systemic infection but not natural infection because this fungus has a limited ability to penetrate insect by the natural route. We have clarified this in the text by adding the sentence: 'The low Toll pathway activation by A. fumigatus is likely due the weak ability of this fungus to penetrate insect by the natural route.'.

3. The use of total KO strains does not preclude the cross talk of cellular and humoral immunity and consequently potential defects in cellular immunity upon deletion of a master regulator of the Toll pathway or even its downstream effectors

The observation that Toll deficient mutants are almost as susceptibility as mutant flies lacking all the four immune modules (△ITPM ) to the five fungal pathogens point to a major role of this pathway. In a previous study (Ryckebusch et al Elife 2025), we have shown that the four immune pathways largely work independently as phagocytosis was still observed in Toll deficient mutant.

4. Did the authors validate that NimC11; Eater1 flies are not able to phagocytose fungal spores?

In the first version of this manuscript, we did not validate that NimC1;eater flies are phagocytic deficient also for Fungal spores although our manuscript assumed it. To address the comment of the reviewer, we have extended our study to better characterize the role of the cellular immune response to fungal infection (See new Figure S1).

Our new results show that NimC1;eater deficient flies have defect in binding to M. anisopliae GFP spores (New Supplement Figure S1E,F). We did not see clear evidence of internalization. Thus, we conclude that the use of NimC1;eater flies is adequate to study the role of the cellular response. We have monitored the survival of hemoless flies that lack nearly all plasmatocytes due to the over-expression of the proapoptotic gene Bax, to natural infection and septic injury with B. bassiana and M. anisopliae. This new piece of data (described in New Supplementary Figure S1A-D) show that hemoless flies display a wild-type survival to B. Bassiana and a mild susceptibility to M. anisopliae consistent with our previous statement that the cellular response is less important than the humoral response. In the revised version, we have added this new piece of data and nuanced our statement on the role of the cellular response to fungal infection.

5. Is it possible that entomopathogenic fungi bypass phagocytosis as a virulence strategy by inducing large size germinating cells, which are not phagocytosed?

Indeed, there are several studies have showed that entomopathogenic fungi have evolved sophisticated strategies to evade or survive phagocytosis.

• Once fungal spores (conidia) germinate, penetrate host tegument and reach the hemocoel, fungi existwithin the hemocoel in the forms of blastospores with thinner cell walls than conidia (M. anisopliae, M. rileyi, B. bassiana), and cell wall-free protoplasts (E. muscae). Wang and St Leger (2006) had demonstrated that host hemocytes can recognize and ingest conidia of M. anisopliae, but this capacity is lost on production of blastospore, because of its ability to avoid detection depending on the cell surface hydrophobic protein gene Mcl1 that is expressed within 20 min of the fungal pathogen contacting hemolymph.
• Other studieshave shown that blastospores of B. bassiana and M. anisopliae can be phagocytosed at the early stages of infection but manage to emerge from host cells and continue to propagate. Growing hyphal bodies can deform the plasmatocyte cell membrane (Gillespie et al., 2000; Hung and Boucias, 1992; Vilcinskas et al., 1997). Studies have also shown that during the infection process of entomopathogenic fungi in insects, the hemocyte count gradually decreases. For instance, during the infection of Thitarodes xiaojinensis by Ophiocordyceps sinensis, blastospores are the initial cell type present in the host hemocoel and remained for 5 months or more before transformation into hypha, which finally led to host death; and the increase in blastospores quantity coincidence with a decline in hemocyte count (Liu et al., 2019; Li et al., 2020).<br /> In a new set of experiments, we tested the ability of plasmatocytes to phagocytose M. anisopliae-GFP spores. We observed that plasmatocytes bind to the spores, but we did not obtain clear evidence of internalization (New Figure S1E,F). However, this assay was not sufficient to conclusively determine whether plasmatocytes internalize M. anisopliae spores, as GFP fluorescence may be quenched in acidic intracellular compartments. Because entomopathogenic fungi can affect hemocyte abundance, we also monitored the expression level of Hml, a hemocyte-specific marker, in flies following natural infection with B. bassiana, M. anisopliae, M. rileyi, and E. muscae at 2, 3, and 5 days post-infection (see figure below). We did not observe a reduction in hemocyte levels for any of these fungi except M. anisopliae. This suggests that M. anisopliae may reduce hemocyte numbers as a strategy to circumvent the cellular immune response. These results, although promising, were not included in the revised version of the manuscript, as a thorough analysis of the cellular immune response would require a dedicated study on its own.

Figure: Expression of Hml by RT-qPCR upon natural infection with entomopathogenic fungi (figure not included in the revised manuscript)

6. Is it possible that fungal toxins kill phagocytes during germination?

There are indeed evidences that fungal toxins destruxins (DTXs) induce ultrastructural alterations of circulating plasmatocytes and sessile haemocytes of Galleria mellonella larvae. DTXs contribute to the fungal infection process by a true immune-inhibitory effect. This is evidenced by two key findings: first, the germination rate of injected Aspergillus niger spores was slightly but significantly enhanced; second, during incubation, the fungus demonstrated a greater ability to escape from the haemocyte-formed granuloma envelope (Vilcinskas et al., 1997; Vey et al., 2002). But in Drosophila, Destruxin does not appear to affect Drosophila cellular immune responses in vivo. Phagocytosis of E. coli bacterial particles in Destruxin-injected flies appeared to be the same as that seen in PBS-injected flies. The proliferation of bacteria in the Destruxin-injected flies was due to the lower expression of antimicrobial peptide genes suggesting that Destruxin A specifically suppressed the humoral immune response in Drosophila (Pal et al., 2007), which is consistent with major role of antimicrobial peptides in survival to fungi. This point is now discussed in the discussion with a new section on the cellular response to fungal infection.

__Reviewer #3 (Significance (Required)):____ __

This is an important work that provide new information on virulence mechanisms of entomopathogenic fungi and the host immune responses that mediate host protection. The authors should address my comments in the discussion and provide some additional evidence by using reporter fungal strains for hemocytes on whether these fungal pathogens completely bypass phagocytosis to invade the host. Therefore, rather than claiming that phagocytosis is not important it should be clarified whether phagocytes are directly involved in host defense or whether the fungus changes its cell wall surface to avoid this line of host defense. My expertise is on phagocyte biology and host-fungal interaction on human fungal pathogens.

We have added more information showing that plasmatocytes of NimC1;eater larvae fail to bind to spores of M. anisopliae suggesting that this line provides an appropriate tool to assess phagocytosis. We have also analyzed the survival of flies depleted for plasmatocytes via the over-expression of bax, which revealed a mild role for plasmatocyte in defense against M. anisopliae but not B. bassiana. By performing additional experiments, we realized that analyzing the role of cellular immunity in host defense against these five fungi would require much more work and is beyond the scope of this study. We have however added in the revised version a para in the discussion on the the cellular response.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #3

Evidence, reproducibility and clarity

This is a clearly written manuscript on the immune effector mechanisms regulating Drosophila melanogaster host defense against a broad range of fungal pathogens, including entomopathogenic and saprophytic filamentous fungi. The authors systematically dissect the contribution of major arms of Drosophila immunity, including cellular and humoral responses and melanization and potential mechanisms of cross talk using genetic tools and reporter lines. They also go into detail to characterize the contribution of upstream activators of these responses by fungal PAMPs and the role of antimicrobial effectors (AMPs) in fly susceptibility.

They conclude for no important role of phagocytosis in host defense. Instead, they find important contributions of Toll pathway mainly through the detection of fungal proteases by Persephone rather than b-glucan detection by GNBP3. They also demonstrate that Toll activation is proportional to the virulence of the fungal pathogen, showing little activation of this response by Aspergillus fumigatus. Finally, they identify melanization as another line of host defense that restricts pathogen dissemination and protects fly from invasive fungal disease. A very interesting part of this study is the identification of a virulence strategy of the obligate fungus Entomophthora muscae, which employs a vegetative development strategy, by making protoplast that avoid immune recognition by masking immunostimulatory cell wall molecules to avoid immune recognition by Toll pathway until the very last stage of invasive growth. Overall, this is a very interesting study on host-pathogen interplay in Drosophila, shedding light onto novel pathogenetic mechanism employed by entomopathogenic fungi to adapt to their hosts.

Major comments for the authors:

1. The use of reporter fungal strains to capture the dynamic interplay of the pathogen and the different arms of the immune system precludes firm conclusions on the contribution of various immune response to infection. This should be emphasized in the discussion
2. The route of infection and the method employed to inject fungal spores has an impact on the effector pathways being activated. For example, pricking introduces spores less efficiently in the hemolymph compared to microinjection. The inoculum size in case of microinjection also has profound impact in understanding the role of cellular and humoral immunity during the infection course. For example, the lack of Toll activation in the natural infection with A. fumigatus does not mean that this pathway is not important in host defense against this pathogen.
3. The use of total KO strains does not preclude the cross talk of cellular and humoral immunity and consequently potential defects in cellular immunity upon deletion of a master regulator of the Toll pathway or even its downstream effectors
4. Did the authors validate that NimC11; Eater1 flies are not able to phagocytose fungal spores?
5. Is it possible that entomopathogenic fungi bypass phagocytosis as a virulence strategy by inducing large size germinating cells, which are not phagocytosed?
6. Is it possible that fungal toxins kill phagocytes during germination?

Significance

This is an important work that provide new information on virulence mechanisms of entomopathogenic fungi and the host immune responses that mediate host protection. The authors should address my comments in the discussion and provide some additional evidence by using reporter fungal strains for hemocytes on whether these fungal pathogens completely bypass phagogytosis to invade the host. Therefore, rather than claiming that phagocytosis is not important it should be clarified whether phagocytes are directly involved in host defense or whether the fungus changes its cell wall surface to avoid this line of host defense. My expertise is on phagocyte biology and host-fungal interaction on human fungal pathogens.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #2

Evidence, reproducibility and clarity

In this work the authors describe the contribution of distinct immune responses in Drosophila melanogaster to systemic and natural infections with 5 fungal species with different lifestyles some being generalists infecting a broad range of insects while others being more specialists or opportunistic. The authors used several well characterized Drosophila mutants of the Toll, Imd, phagocytosis and melanization responses to address this question. They show that Toll pathway is the key player in anti-fungal resistance in both natural and septic infections, whereas melanization plays a minor role mainly during natural infections possibly to limit fungal invasion through the cuticle. The authors show elegantly using different combinations of mutants for antimicrobial peptides genes with antifungal activities that Bomanins and Daisho (1 and 2) are the main Toll effectors mediating resistance to fungi but the authors did not find specific fungus-by-gene interaction, but rather antifungal peptides seem to act in a more general fashion against the fungi tested with significant redundancies between certain classes. Interestingly the authors show that while generalists like Beauveria and Metarhizium strongly activate the Toll pathway, the specialist E. muscae weakly activates the pathway and the opportunistic A. fumigatus does not activate the pathway, indicating that certain fungal species are able to evade sensing by immune pathways. In the context of the Toll activation, the sensor protease Psh and not GNBP3 seem to be the main trigger of the pathway.

Minor comments

This is an interesting work that compares the contributions of different arms of the fly immune response to 5 fungal species with diverse lifestyles. The use of different lines with different combinations of mutant genes is a strength to highlight the relative contribution of each immune response. Some of the data obtained is intriguing and warrants more future investigations such as the distinct phenotypes of ModSp and GNBP3 mutants in E. muscae infections. The methodology is robust and the conclusions are supported with good experimental evidence. I do not see any major concerns with the work. I just have some minor comments listed below

1. Statistical significance should be indicated on Figures 1 and 2, although it appears in the legend.
2. It is not very accurate to use the term resistance of the different mutants to infections with the diverse fungal species in Figures 1 and 2 especially that the authors have reported only survival data in these figures and have not measured fungal proliferation in infected flies (although they did that in later figures). It is more accurate to mention that the mutants flies have different levels of tolerance rather than resistance to fungal infections.
3. The authors show that Toll is over-activated in PPO1/PPO2 double mutant possibly through a negative feedback mechanism. However, there could be another explanation for this observation: For instance, the increased fungal proliferation in the PPO double mutant results in increased protease secretion by fungi enhancing Psh activation! Also, how can fungi manage to proliferate in this double mutant if Toll is overactivated? Could it be that Toll overactivation is triggering a fitness cost?
4. In Lines 654-655, it is not accurate to say that E. muscae protoplasts are not detected by the immune response since E. muscae natural infections triggers Drs expression at 24 hpi and there is possibly some melanization taking place since PPO1 and PPO2 are required for defense against this fungus. A more accurate explanation is that this fungus is possibly more resistant to the effectors of the host immune response than the other fungi. I think a major point that the authors might have missed to consider in the discussion of their data is that the different fungi used herein may exhibit different levels of resilience to the effector reactions of the host such as AMPs and melanin deposition

Significance

Although the importance of Toll pathway and melanization in antifungal immunity is not new per se, this work adds to this knowledge by showing that Toll has the upper hand in anti-fungal immunity and that the strength of Toll pathway activation and its effector capacity may vary depending on the type of invading fungus. The work also highlights that certain fungi may employ a delayed switch to hyphal growth to reduce the presence of cell wall sugars as a mechanism to evade immune recognition. Overall, this work significantly adds to the knowledge of Drosophila immunity and raises some interesting questions related to the evolution of host-pathogen interactions and to the complex functions of serine protease cascades regulating Toll and melanization. This work will be of interest to a broad audience in the field of host-pathogen interactions

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #1

Evidence, reproducibility and clarity

The key conclusions are solid. All the claims are supported by quality data. The content is rich, and no additional experiment is needed. The data and methods are properly presented for reproduction. The experiments are adequately replicated. One comment on statistical analysis is listed below.

Summary:

This manuscript investigates how Drosophila immune pathways contribute to defense against a range of filamentous fungi with distinct ecological strategies. The work providesovel insights into Toll pathway activation through pattern recognition receptors and danger signals, relative roles of melanization, phagocytosis, and effects of antimicrobial peptides, and particularly the immune evasion strategy of E. muscae via protoplast formation. These findings are of broad relevance to insect immunology, host-pathogen interactions, and evolutionary biology. The study is well designed, the experiments are carefully executed, and the manuscript is clearly written. It is novel to demonstrate that E. muscae evades immune recognition via protoplast formation. However, some aspects of clarity and discussion of limitations could be improved before publication.

Major comments:

1. The Abstract is informative but a bit too long. Consider condensing some sentences and highlighting the novel contributions (e.g., role of protoplasts in immune evasion.).
2. The Results may use more mechanistic links. For instance, the section on E. muscae immune evasion could more explicitly connect the morphological findings (protoplasts, lack of cell wall) with specific immune recognition failures.
3. Please clarify statistical analyses used for survival data (e.g., log-rank tests, multiple testing corrections).

Minor comments:

Abstract: 1) "The infection outcome depends on the complex interplay between insect immune defenses and fungal adaptive strategies." could be simplified to: "Infection outcomes depend on the interplay between insect immunity and fungal adaptation." 2) Replace "our study uncovers" with "we show" for more concise phrasing. Reduce phrases like "our study reveals" or 'we conclude" in other parts of the manuscript. Results: p. 5: phrase "survival upon natural infection... reveals the major contribution" → reword to avoid passive tone. p. 10: clarify "vesicles push the membrane outwards" with more precise terminology (e.g., budding, extrusion). Discussion: p. 20: streamline sentence beginning "These observations provide a mechanistic basis..." (currently too dense).

Referee cross-commenting

I agree with the comments of the other two reviewers.

Significance

This manuscript investigates how Drosophila immune pathways contribute to defense against a range of filamentous fungi with distinct ecological strategies (generalists, specialists, opportunists). By leveraging a comprehensive panel of genetically defined fly lines and standardized infections, the authors provide a demonstration that the Toll pathway is the predominant systemic antifungal defense, extending classical findings into a comparative framework across fungal lifestyles. The work provides novel insights into Toll pathway activation through GNBP3 and fungal proteases sensed by Psh, while also dissecting the relative contributions of melanization, phagocytosis, and antimicrobial peptides to host protection. Of particular note is the compelling demonstration that the fly specialist E. muscae can evade immune recognition through protoplast-like vegetative forms, minimizing cell-wall exposure and thereby escaping Toll activation.

My expertise and limitations:

Insect biochemistry and molecular biology, with particular focus on innate immunity, serine protease cascades, melanization, and host-pathogen interactions. I also have experience with genetic, biochemical, and functional approaches to dissecting immune signaling pathways in model insects. However, I do not have sufficient expertise to critically evaluate advanced statistical analyses.

Because a LLM doesn't have any intelligence/cognitive ability or agency to "apologize", let alone "remember" it.

That's why you shouldn't pick complexities. World in not black and white. Progress is not linear.

Hah, you mention Iran from the Gulf and not Saudi Arabia? Also, besides Russia and Saudi Arabia where it's oligarchs, most of this goes to companies, and corrupt politicians are bought or invisibilised. Also, the US has the largest fracking soil in the world...

Reminds me of McLuhan's tetrad!

Now it's dead, because flash is dead.

Are emojis processed like words?: Eye movements reveal the time course of semantic processing for emojified text in Zotero

[[Reading in the Brain by Stanislas Dehaene]] https://archive.org/details/readinginbrainne0000deha

Are emojis processed like words?: Eye movements reveal the time course of semantic processing for emojified text in Zotero

Emojis in text clash with reading (two different modes of processing) esp when the emoji are incongruent.

The Ss do this as I will be playing its audio in T B.

The page relies heavily on scrolling and pointer-based interaction to reveal information across wide, horizontally arranged sections. This interaction style may be challenging for keyboard-only users or those using assistive technologies if alternative navigation options are limited.

Clear and consistent headings help users understand the structure of the page. This is especially important for screen reader users who rely on headings to navigate content efficiently.

Apple uses strong contrast between text and background, making content readable for users with low vision or colour blindness. This supports the POUR principle of being perceivable.

nil

nil

nil

nil

Nil

Nil

I appreciate the options for meetings with consultants because as someone will a very busy and full schedule this semester, I could use this type of flexibility.

I like this because it can give students a very good outline for what an assignment should look like. I think this is especially good for an online class since we do not see our professor in person to ask questions.

This is important for me to remember because I tend to copy my things and paste them with other ideas to compare what I like more, I will remember this for the purpose of this class

I will have to remember to do this. Sometimes I forget little things such as this so I will make note of this!

I might be a little silly for this question, but why is that? Does the process of copying and pasting a word document into google docs do something? Or does it make the system think it was AI? I'm genuinely curious about this one. Also glad to be using docs instead of word, word gets too glitchy after writing in it for a bit.

This is really understandable, especially because this is a writing class about our own thoughts. Not what a robot thinks. So, i believe the harshness pushed through the syllabus to really enforce this is necessary.

Thank you for adding this, because sometimes I do email professors in the evening and nights when they aren’t available to get back to me right away during those times. I will keep in mind to plan ahead when I do have questions and comments.

This! I do want to be a children’s author but I feel as though I am so unfamiliar with the perspective of writing and fiction. However, I strongly want to go into the fiction writing category. I really do hope that I would learn to become familiar with fiction and prose and start to write my own by the end of this course.

Good reminder to take advantage of writing centers provided by the school. As someone who loves to write but have a hard time with grammar and writing efficiently on some aspects, the writing center could definitely help me immensely in my writing process and organize my ideas.

I’m honestly glad that AI is not allowed in this creative writing class. I believe it takes away from the fun and creativity of students. Especially in today’s time, AI has taken over many things and I just hope it doesn’t take over the learning process for any academic schooling. Yes, it makes it easier but it isn’t going to benefit students in the long term, because it causes students to cheat and not rely on their cognition as much. It would be pointless, because you wouldn’t gain anything from the class you are paying for.

Google docs is so much easier for me to use. I just started using it more this year and I think its helpful for not only me but the professors. It is easier for me to use it through canvas as well!

This is so helpful. Having the readings/textbook online works so much easier for me personally because I can do my work and have the readings pulled up right next to it on my laptop!

I did not know that this was an option! I think that is a really helpful option just in case if something were to happen. Im glad this is in the syllabus because i would never have known this prior to reading this.

I really like that there is a flexible extension option. I think this is fair and gives students the option to take accountability when an assignment is late or will be late.

I really appreciate this because I am always very hard on myself about my writing. I try my best but it doesn't always turn out the way I want it to. I'm slowly learning that most people don't see as many flaws in it as I do.

I like that this is integrated into the course. Usually in my previous classes I haven't been giving the opportunity to learn and practice my skills without it being worth a big percentage of my grade.

I am happy to see this was included in the syllabus! Especailly with the state the government is in with taking away food stamps etc.

I completely agree and this is a very fair outcome. AI is dimishing and harmful to use as students pursuing a bachelors degree. Not only does it harm the evniornment and negatively impacts the work force. It also impacts student learning and defeats the purpose of going to unviersity in the first place.

I honestly appreciate this. Some courses don't allow room for late work. I am super grateful that this course has some flexiblity built into it, allowing for flexbility as well as accountability as students enrolled in this course.

I had no idea that they offered online and email feedback! This was genuinely helpful information, as I've always felt somewhat nervous to go to the writing center. Thank you for including this!

It saddens me that this is something that has to be stated, in a writing class of all things. I can understand the reasoning and rationalization some would use, but in a course like this, it only hinders you to use AI. It's also disrespectful to your professor and peers.

Honestly, I really appreciate this set of expectations being laid out to begin. Some of my professors have been quite vague in their grading policy, or refuse to give anyone a 100% because "nobody can be perfect." In an English class of all things, when it's such an subjective territory, this is very considerate of you.

Throughout my experience reading various works of fiction, I have not been able to examine the craft of writing it. I am incredibly excited to see what this class has in store for me!

Rereading things is very important for this exact reason. Once you know the general theme and lay out of what you are reading, you are able to recognize and notice more little things you may have missed the first time.

This is good to remember not only about others but in a personal sense as well. Sometimes being academically smart seems like the most important but there are many ways to be smart. Some of the people in my life I consider the smartest tend to have many other areas of intelligence, not just in academics.

These questions are important to pay attention to because it can have an affect on how you interpret the essay you are reading. Knowing more about the author and the process of the essay may even give you better perspective on material.

This is a good example of an annotation. Reflect and ask yourself about the information.

It is good to reflect on what you read before forgetting it right away. Slowing down and reflecting will be more efficient.

These steps can be very helpful in understanding the information being read. Thoroughly following them will be beneficial.

I actually never thought about it like that, In book club in high school. I'd be drawn to some books, but others i could barely pick up. Never knew the reason why or how to explain it. Wish I was taught to actively read more then, cause then I would be more aware of it all.

I think taking it all in in one breath is really effective as someone who has bad memory. If I took a pause somewhere during my reading, majority of the time, I have to reread it again so reading everything in one setting will save time. If I really do need to pause, leaving a detailed annotation would help me see where I left off and what I am reading about so far.

This annotation is short and simple, yet it explains the main idea of the whole paragraph. I hope to actually start having a simple, yet effective annotation.

Good idea. I do seem to be on the timid side and find myself under annotating a text and sometimes over annotating. I think sometimes a word or phrase is just as good as a sentence like stated. When I over annotate, it is harder to go back and search through important context and I find myself being lost.

I noticed, that when I do choose a piece I want to read, the title is the first thing that really captures my eyes. When a title is interesting, I believe it makes you want to dive deeper to see what the writing is really about.

There were many instances in which I had to reread lines often when I read to make sure I’m getting the point the writer is trying to convey. Writers aren’t going to be the same so active reading and becoming familiar with their writing style is helpful when processing and analyzing what I’ve just read.

This guidance pushed me to approach the reading in a more active and careful way. Asking simple questions about the essay’s meaning and organization helped me better grasp the author’s ideas. As a result, I stayed more focused and engaged with the text.

This reminder helped me feel more comfortable while reading because it showed that annotating is a flexible process. I focused on the parts of the text that stood out to me or caused questions, rather than trying to mark everything. This made my annotations more meaningful and useful for understanding the essay.

As I read the essay, I noticed that it presented a mix of familiar and unexpected ideas that challenged my initial understanding of the writer’s message. Some arguments caused me to rethink the author’s purpose as I moved through the text. I also paid attention to unfamiliar words so I could look them up later and better understand the essay as a whole.

Ive always struggled more with reading something twice especially when its a long reading that I am not particularly interested in.

This will be very helpful for me when reading a new selection. Having steps and actually incorporating them into reading will be very beneficial for me in this course.

I understand why this is necessary but I rarely reread passages unless I wasn't paying attention the first time around. I have however read scripts multiple times and done research on the language used in them. I think it's very important as a performer to really understand what you are sharing with an audience. If it doesn't make sense to you it almost surely will not make sense to them!!

We talked about this a lot in my American Lit. class last semester. It's really interesting how writers incorporate real life situations and ideals into their fiction writing whether on purpose or accidentally. You can tell a lot about a writer through their work when you know about their background. For example, Mark Twain's Adventures of Huckleberry Finn. Mark Twain was very critical of religion and romanticism. He went on to include a lot of irony regarding those things in his work!

We're covering something like this in my mass media class. It's important to analyze how a piece of literature makes you feel. Sometimes they way a reader interprets a piece is not exactly what the author intended which is really interesting. You should really be thinking about what the author is saying and why.

This is a very important skill. I often read things and stop in the middle because I don't understand something which really disrupts the flow of what I'm reading.

I have found this to be a very helpful skill. If by the second read I still have question in my notes it makes it much easier to have them written down so I don't forget them. I can then ask for help or come back to the question later.

I believe this is one of the most important habits to develop. The second read really allows you to understand the text and I have found it very helpful for scientific articles.

Some of the smartest people I know are lifelong learners. They are more concerned with what they do not know and learning new things rather than becoming fixated on their area of knowledge only.

There are multiple different ways to be intellegent. Someone can be street smart, book smart, business savy. Everyone has their own strengths and things that they excel at. Things like IQ tests and standarized testing isn't an accurate way to asses someone's intellegence. I wish they would try to figure out another way other than streneous exams.

I honestly think I am a very lazy reader. I usually just read the article once and am left confused. I can definietly see how reading the article a second time allows for more appreciation of the material.

Taking marginal notes is super clever! I was never taught or thought of the idea myself to do that. I can see how that definetly helps further your understanding and whatt he passage is actually conveying.

I am eager to start actively reading as a STEM major, I often find myself struggling to truly understand readings and find a major takeaway. Hopefully I can interact with more of my readings to better absorb the content.

I think assessment of the reputation of the publisher is vital here as well, not just with the author. This can provide context on what kinds of work the publisher typically endorses.

This is crucial and aids in the idea of developing critical thinking skills. It's something a lot of people forget when it comes to research articles, but it's relevant to. Being able to thoroughly assess the position of an author and any conflicts of interests they may have that might influence said opinion important when weighing their thoughts,

I can certainly attest to this. I'm currently in Intro to Women's Literature, and we read a sample of poems from different points and places in history. When reading them, I was completely lost at points, as valuable context about what environment the authors lived in was largely missing. When the authors referred to names and locations, I had no idea what it was supposed to mean, as I had no context.

This is a valuable point to add. Writing is built on logic and validity, whether it be grammatical correctness (according to whatever dominant idea of language there is), the structure of an essay, or the way you introduce arguments.

I like the idea of mentally separating the first and the second reading. It gives the reader a chance to process the content and then go back to examine the author's intent later, once they have a better understanding of the work itself.

This is also an important step in seeing how authors can change your viewpoints or surprise you throughout their works. I also find that this is helpful to the reader when trying to retrace their thought process and determine how or why the author encouraged us to think a certain way.

I think that a big part of being able to read a wide variety of works is not necessarily liking or even agreeing with them, but instead being able to find curiousity in why authors thought or wrote the way they did. I believe this is also something big when studying classical literature, as I find the thought behind them is often more entertaining than the plots.

I got to climb the Pyramid of Cholula which is the biggest pyramid in volume in the world, and it was very difficult to climb albeit not being as tall as the Pyramid of the Sun. I could only imagine having to climb these pyramids instead on a scorching hot day!

Many people seem to think that chocolate started somewhere in Europe, but I'd like to highlight the fact that it actually came from Indigenous cultures in what is now present-day Mexico. It's super important to acknowledge this fact because of just how culturally significant it was for Mesoamericans. Present-day Mexico still makes chocolate that stands out from all competitors. Trust me on this; try a Mexican hot chocolate mix named Chocolate Abuelita. It will absolutely change your view on how chocolate should taste.

This section shows that Chat GPT is not the only AI site and that none of them should ever be used. Your own work is the best work.

This section will help with creating a good week by week routine and complete the work on time.

It is important to remember this section for writing assignments for that you can earn the points for it.

I really like how in this class you do not do A-. I think they are stupid because it should be worth a 4.0. But I like how you also inculde 3.3 for the B-. That is super cool so thank you.

I also really appreciate this as well! I appreciate him allowing the class to feel at home and feel included. Very respectful.

I see pros and cons of academic freedom. I feel like every teacher wants it, and good teachers would thrive with it; however, it is so important to align as a grade level and vertically with other grade level expectations. If you are not on the same page, the students will suffer the following year.

This is infuriating and should be a conflict of interest that is not allowed. Is this legal now days?

I am sure this is still true today but I am curious to investigate further to find examples.

I had not written the annotation I wanted to, because I was going to read on to see if it was addressed and this is starting to address it. My question up above when she stated that 'a new curriculum was achieved, is how do you do this when there is so much pressure to use the provided resources with fidelity? I am curious to read on and see how that was addressed.

STUDENT CHOICE! Yes!!

Student buy-in is so critical to success. Unmotivated children are less likely to get what they should from the lessons. I love that this new found motivation was able to be achieved.

This is so important! What is curriculum if it is not helping our students (all) grow?

I love the self awareness and ownership Mary had within her learning and how to challenge herself in a way that was beneficial to her.

PREFORMATIONIST VIEW

CHILDREN BELIEVED TO HAVE: * SENSORY CAPABILITIES * EMOTIONS * MENTAL APTITUDE AT BIRTH

ENVIRONMENT WAS THOUGHT TO PLAY NO ROLE IN DEV.

NATURE VS NURTURE

On page 360-, we see the discussion about Miss Jane Austen and how her reading isn't forceful; it's different. He continues on about how you can't predict the ending of her play, and everything naturally falls into place. It doesn't have a bunch of random side plots cohering into one. Everything is easy to read, easy to understand, and is fulfilling. Her stories have 'real' reasoning and while we think all authors publish books or directors with films have reasoning. During her era, this was amazing. They weren't only entertainment but captivating and set ideas of life in bigger ways.

From my understanding of page 353, the novel was breaking into the dreamier aspect and away from the realities of life. It made people unsure of how to feel about this new type of fiction. I think it's kind of funny to picture because to me personally I feel like men see it as the way standards were being raised and how they'd have to act to swoon someone. However, maybe I'm misunderstanding its true meaning. It's very interesting how the more modern writing got the more uneasy people begin to be.

Whatley is quite the perplexing individual in that he validates the critiques of others while subtly asserts them as blemishes. It is bold and at the same time rather unapologetic in that manner; however, now I wonder what other critics were saying about Jane Austin during her time. It is clear that Whatley holds her works to a high regard and seems to be the apotheosis of literature from his perspective.

SIDE NOTE: Based on the claim right afterwards, I think while reading Jane Austin's works I will be paying closer attention to characters and how they are utilized.

It is fascinating that Whatley is commending Jane Austin for covert didactic narratives. To have the moral lessons being in arm's grasp, if only, the individual seeks it out. Though, if I was to critique this rationale, to suggest that this methodology is "conformed more closely to real life" not evidently true. Often times, the big moral lessons in real life become loud and clear with great consequences. Beyond the critique, it is fascinating that Jane Austen wrote works with social realism predating the Realism literary period. Revolutionary and "novel" in application.

SIDE NOTE: I hope Whatley lived long enough to see the emergence of the Realism literary period. Given this was written in 1818, I hope he lived 22 years to at least see prevalence of Realism as that seems to be his fascination.

I think what the passage is trying to explain is that the moral lessons in Jane Austen’s novels are delivered with subtlety and grace. They are expressed clearly and powerfully, and never in a way that feels pushy or preachy. Instead of being imposed on the reader, the lessons in the novels happen naturally from the events and situations within the stories, and as a result, the reader can easily understand the intended message on their own.

This message helps my boss to understanding the situation before the morning so that they are not rushing or interrupted in the morning. Furthermore, they might be able to find someone to help with what was once my task.

The receiver (my boss) needs to formally know that I am ill and I will not be able to make it to the meeting. They also need to know what I have already done and when I will be able to get back to work again.

Early programmers had to work directly in binary, which meant they were literally writing instructions in the same 1s and 0s that the computer understood. It's kind of wild to think about having to manually translate everything you want the computer to do into that low-level language without any of the shortcuts we have now.

This is basically saying that living virtuously isn't just about doing the right thing once, but about building good habits that make your life better overall. The idea that virtue gets easier with practice makes sense. It's like how any skill improves the more you work at it.

There are also bots that push back against companies or organizations in ways that might actually be justified. Like there's this "Gender Pay Gap Bot" on Twitter that's hooked up to a database of UK companies' pay gaps, and on International Women's Day it automatically calls out any of those companies that tweet about celebrating the day by quote-tweeting them with their actual pay gap numbers.

LATE ADULTHOOD

2 CATEGORIES 65-84 THE YOUNG-OLD 85+ OLDEST-OLD

EARLY CHILDHOOD

"PRESCHOOL YEARS"

PRENATAL DEV CONCEPTION--DEV. BEGINS- MAJOR STRUCTURES OF THE BODY ARE FORMING WITH THE HELP OF THE MOTHER

VERY similar to nazism

Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

Learn more at Review Commons

Reply to the reviewers

Response to Reviewer 1:

The authors introduce G2PT, a hierarchical graph transformer model that integrates genetic variants (SNPs), gene annotations, and multigenic systems (Gene Ontology) to predict and interpret complex traits.

We thank the reviewer for this accurate summary of our approach and contributions.

Major Comments:

Comment 1-1. Insufficient Specification of Model Architecture: The description of the "hierarchical graph transformer" lacks technical depth. Key implementation details are missing: how node embeddings are initialized for SNPs, genes, and systems; how graph connectivity is defined at each level (e.g., adjacency matrices used in Equations 5-9, the sparsity); justification for the choice of embedding dimension and number of attention heads, including any sensitivity analysis; and the architecture of the feed-forward neural networks (e.g., number of layers, activation functions, and hidden dimensions).

__Reply 1-1. __As requested, we have expanded the technical description of the model architecture, including the hierarchical graph transformer (HiGT), in the Materials and Methods section. Details regarding node initialization and hierarchical connectivity are now included in the new paragraph "Model Initialization and Graph Construction." Specifically, all node embeddings corresponding to SNPs, genes, and ontology-defined systems are initialized using uniform Xavier initialization (Glorot and Bengio, 2010).

We have also clarified our hyperparameter optimization strategy. Learning rate, weight decay, hidden (embedding) dimension, and the number of attention heads were selected via grid search, as summarized in new Supplementary Fig. 8, reproduced below. Based on both performance and computational efficiency, we adopted four attention heads-consistent with the configuration commonly used in academic transformer models (Vaswani et al., 2017) (the original Transformer used eight).

Regarding the feed-forward neural network, we follow the standard Transformer architecture consisting of two position-wise layers with hidden dimension four times larger than the node embedding size and a GeLU nonlinear activation function (Hendrycks and Gimpel, 2016). This configuration is widely established in the literature and functions as an intermediate processing step following attention; therefore, it is not a focus of hyperparameter tuning. All corresponding updates have been incorporated into the revised Methods section for clarity and completeness.

Comment 1-2. No Simulation Studies to Validate Epistasis Detection: The ground truth epistasis interaction should use the ones that have been manually validated by literature. The central claim of discovering epistatic interactions relies heavily on the model's attention mechanism and downstream statistical filtering. However, no simulation studies are presented to validate that G2PT can reliably detect epistasis when ground-truth interactions are known. Demonstrating robust detection of non-additive interactions under varying genetic architectures and noise levels in simulated genotype-phenotype datasets is essential to substantiate the method's core capability.

Reply 1-2. We agree that a simulation of epistasis detection using the G2PT model is a worthy addition to the manuscript. Accordingly, we have now incorporated a new section in the Results titled "Validation of Epistasis through Simulation Studies", which includes two new figures reproduced below (Supplementary Fig. 6 and Fig. 5). We have also added a new Methods section to describe this simulation study under the heading "Epistasis Simulation". These simulation studies show that G2PT recovers epistatic gene pairs with high fidelity when these pairs are coherent with the systems ontology (c.f. 'ontology coherence' in Supplementary Fig. 6, which reflects the probability that both SNPs are assigned to the same leaf system). Furthermore, G2PT outcompetes previous tools, such as PLINK-epistasis, which do not use knowledge of the systems hierarchy in the same way (Supplementary Fig 6b-d). Using simulation parameters consistent with current genome-wide association studies (n = 400,000) and understanding of heritability (h2 = 0.3 to 0.5) (Bloom et al. 2015; Speed and Evans 2023), we find that approximately 10% of all epistatic SNP pairs can be recovered at a precision of 50% (Fig. 5). We have provided the source code for this simulation study in our GitHub repository (https://github.com/idekerlab/G2PT/blob/master/Epistasis_simulation.ipynb)

Comment 1-3. Lack of Justification for Model Complexity and Missing Ablation Insights: While Supplementary Figure 2 presents ablation studies, the manuscript needs to justify the high computational cost (168 GPU hours using 4×A30 GPUs) of the full model. It remains unclear how much performance gain is specifically due to reverse propagation (Equations 8-9), which is claimed to capture biological context. The benefit of using a full Gene Ontology hierarchy versus a flat system list is not quantified. There is also no comparison between bidirectional versus unidirectional propagation. Overall, the added complexity is not empirically shown to be necessary

Reply 1-3. We thank the reviewer for prompting a clearer justification of complexity and ablations. We have now revised the Results to (i) quantify the specific value of the ontology and reverse propagation, and (ii) explain why a flat SNP→system model is computationally and biologically sub-optimal. We have added new ablation results to compare bidirectional (forward+reverse) versus forward-only propagation. Reverse propagation has little effect when epistatic pairs are within one system (ontology coherence ρ=1.0) but substantially improves retrieval when interactions span related systems (e.g., ρ≈0.8) (Figure reproduced below) A flat design scores a dense genes×systems map, ignoring known sparsity (sparse SNP→gene assignments; sparse ontology edges) and losing multi-scale context; our hierarchical formulation restricts computation to observed edges (SNP→gene→system) and aggregates signals across levels, yielding better efficiency and biological fidelity.

Comment 1-4. Non-Equivalent Benchmarking Against PRS Methods: Figure 2 compares G2PT to polygenic risk score (PRS) methods such as LDpred2 and Lassosum, but G2PT is run only on SNPs pre-filtered by marginal association (p-values between 10⁻⁵ and 10⁻⁸), while the PRS methods use genome-wide SNPs. This introduces a strong bias in G2PT's favor by effectively removing noise. A fair comparison would require: (a) running LDpred2 and Lassosum on the same pre-filtered SNP sets as G2PT, or (b) running G2PT on genome-wide or LD-pruned SNP sets. The reported superior performance of G2PT may be driven primarily by this input filtering, not the model architecture.

Reply 1-4. We appreciate the reviewer's concern regarding benchmarking equivalence. In response, we have extended our analyses to include PRS-CS (Ge et al., 2019) and SBayesRC (Zheng et al., 2024), two state-of-the-art Bayesian shrinkage methods comparable to LDpred2 and Lassosum. Although we initially attempted to run LDpred2 and Lassosum under all SNP-filtering conditions, their computational requirements at UK Biobank scale proved prohibitively time consuming. We therefore focused on PRS-CS and SBayesRC, which offer similar modeling principles with greater computational tractability. These methods have now been run at matched SNP-filtering conditions to our original study. The new results demonstrate that G2PT consistently outperforms PRS-CS and SBayesRC (new Fig. 2, reproduced below), indicating that its performance advantage is not solely attributable to SNP pre-filtering but also to its hierarchical attention-based architecture.

Comment 1-5: No Details on Hyperparameter Optimization: Although the manuscript mentions grid search for hyperparameter tuning, it provides no information about which parameters were optimized (e.g., learning rate, dropout rate, weight decay, attention dropout, FFNN dimensions), what search space was explored, or what final values were selected. There is also no assessment of how sensitive the model's performance is to these choices. Better transparency would help facilitate reproducibility

Reply 1-5. We agree with the reviewer and have expanded the manuscript to include full details of hyperparameter optimization. As described in the revised Methods section, we performed a grid search over learning rate {10−3,10−4,10−5} hidden dimension {64,128} and weight decay {0,10−5,10−3}. The results, summarized in Supplementary Fig. 8 (reproduced above), show that model performance is most sensitive to the learning rate, while hidden dimension and weight decay exert more moderate effects. Based on these findings, we selected a learning rate of 10−5, hidden dimension of 64, and weight decay of 10−3 for all subsequent experiments. Although a hidden dimension of 128 slightly improved performance, we adopted 64 to balance predictive accuracy with computational efficiency.

Comment 1-6. Absence of Control for Key Confounders: In interpreting attention scores as reflecting genetic relevance (e.g., the role of the immunoglobulin system), the model includes only age, sex, and genetic principal components as covariates. Important confounders such as BMI, alcohol use, or medication (e.g., statins) have not been controlled for. Since TG/HDL levels are strongly influenced by environment and lifestyle, it is entirely plausible that some high-attention features reflect environmental tagging, not biological causality.

Reply 1-6. In the current framework, we included age, sex, and genetic principal components to account for demographic and population-structure effects, focusing on genetic contributions within a controlled baseline. We acknowledge that non-genetic covariates can influence downstream biological states and may indirectly shape attention at the gene or system level. Accurately modeling such effects requires an extended framework where environmental variables directly modulate gene and system embeddings rather than being implicitly absorbed by the attention mechanism. We have clarified these limitations in the Discussion along with plans to incorporate explicit confounder modeling in future extensions of G2PT.

Comment 1-7. Oversimplified Treatment of SNP-to-Gene Mapping: The SNP-to-gene mapping strategy combines cS2G, eQTL, and nearest-gene annotations, but the limitations of this approach are not adequately addressed. The manuscript does not specify how conflicts between methods are resolved or what fraction of SNPs map ambiguously to multiple genes. Supplementary Figure 2 shows model performance degrades when using only nearest-gene mapping, but there is no systematic analysis of how mapping uncertainties propagate through the hierarchy and affect attention or interpretation.

Reply 1-7. In the revision (Results), we have clarified how conflicts between cS2G, eQTL, and nearest-gene annotations are resolved, and we have reported the proportion of SNPs that map to multiple genes across these three annotation approaches. We note that the hierarchical attention mechanism enables the model to prioritize among alternative gene mappings in a data-driven manner, and this is a major strength of the approach. As shown in Fig. 3 (Results, reproduced below), SNP-to-gene attention weights reveal dominant linkages, reducing the impact of mapping uncertainty on interpretation. We now explicitly describe this mechanism and acknowledge that further work in probabilistic mapping and fine-mapping approaches is a valuable future direction for improving resolution and interpretability.

"For SNPs with several potential SNP-to-gene mappings (Methods), we found that G2PT often prioritized one of these genes in particular due to its membership in a high-attention system. For example, the chr11q23.3 locus contains multiple genes including the APOA1/C3/A4/A5 gene cluster (Fig. 3c) which is well-known to govern lipid transport, an important system for G2PT predictions (Fig. 3a). Due to high linkage disequilibrium in the region, all of its associated SNPs had multiple alternative gene mappings available. For example, SNP rs1145189 mapped not only to APOA5 but to the more proximal BUD13, a gene functioning in spliceosomal assembly (a system receiving substantially lower G2PT attention). Here, the relevant information flow learned by G2PT was from rs1145189 to APOA5 to lipid transport and protein-lipid complex remodeling (Fig. 3c; and conversely, deprioritizing BUD13 as an effector gene for TG/HDL). We found that this particular genetic flow was corroborated by exome sequencing, which implicates APOA5 but not BUD13 in regulation of TG/HDL, using data that were not available to G2PT. Similarly, two other SNPs at this locus - rs518547 and rs11216169 - had potential mappings to their closest gene SIK3, where they reside within an intron, but also to regulatory elements for the more distant lipid transport genes APOC3 and APOA4. Here, G2PT preferentially weighted the mappings to APOC3 and APOA4 rather than to SIK3 (Fig. 3c)."

Comment 1-8. Naive Scoring of System Importance: The method used to quantify the biological relevance of systems (i.e., correlating attention scores with predicted phenotype values) risks circular reasoning. Since the model is trained to optimize prediction, systems that contribute strongly to prediction will naturally show high correlation-even if they are not biologically causal. No comparison is made with established gene set enrichment methods applied to GWAS summary statistics. The approach lacks an independent benchmark to validate that the "important" systems are biologically meaningful.

Reply 1-8. As requested, we compared G2PT's system-level importance scores with results from MAGMA competitive gene-set analysis, an established enrichment approach. This analysis indeed shows significant correlation between the systems identified by the two approaches (ρ = 0.26, p .01; Supplementary Table. 2), reflecting a shared emphasis on canonical lipid processes. We also observed systems detected by G2PT but not strongly detected by MAGMA's linear enrichment model-for example, the lipopolysaccharide-mediated signaling pathway (Kalita et al. 2022)

Comment 1-9. No External Validation to Assess Generalizability. All evaluations are performed using cross-validation within the UK Biobank. There is no assessment of generalizability to independent cohorts or diverse ancestries. Given population structure, genotyping platform, and phenotype measurement variability, external validation is essential before claiming the method is suitable for broader use in polygenic risk assessment.

Reply 1-9. To externally validate the G2PT model requires individual level genotype data with paired TG/HDL measurements, sample size at the scale of the UK Biobank, and GPU access to this data. Thus, we approached the All of Us program, a large and diverse cohort with individual level data and T2D conditions with HbA1C measurements. We first processed the All of Us genotype and phenotype data as we had processed UKBB data (Methods), resulting in 41,849 participants with T2D and 80,491 without T2D across various ethnicities. We then transferred the trained T2D G2PT model to the AoU Workbench and evaluated its performance. The model demonstrated robust discriminative capability with an explained variance of 0.025, as shown in the new Fig. 2d, (reproduced above).

Comment 1-10. Computational Burden and Scalability Are Not Addressed: The paper notes that training the model requires 168 GPU hours on 4×A30 GPUs for just ~5,000 SNPs. However, there is no discussion of whether G2PT can scale to larger SNP sets (e.g., genome-wide imputed data) or more complex biological hierarchies (e.g., Reactome pathways). Without addressing scalability, the model's applicability to real-world, large-scale genomic datasets remains unclear.

Reply 1-10. We have addressed scalability with both engineering optimizations and new scalability experiments. First, we refactored the model to use the xFormer memory-efficient attention for the hierarchical graph transformer (Lefaudeux et al., 2022), which also helps full parallelization of training, reducing bottlenecks. Second, we added a scaling study with progressively increasing SNP count. On 4×A30 GPUs, end-to-end training time for the 5k-SNP setting decreased from 4000 to 400 min. (approximately 7 GPU-hours, ×10). These new results are given in Supplementary Fig. 7, reproduced below.

Minor Comment:

Comment 1-11. Attention Weights as Mechanistic Insight: The paper equates high attention scores with biological importance, for example in highlighting the immunoglobulin system. There is no causal validation showing that altering the highlighted SNPs, genes, or systems has an actual effect on TG/HDL. Attention weights in transformer models are known to sometimes reflect spurious correlations, especially in high-dimensional settings. The correlation between attention scores and predictions (Supplementary Fig. 3a,b) does not constitute biological evidence. The interpretability claims can be restated without supporting functional or causal validation.

Reply 1-11. We thank the reviewer for this thoughtful comment. We agree that attention weights are not causal evidence. In the revision, we (1) reframe attention-based findings as hypothesis-generating rather than mechanistic, and (2) add an explicit limitation noting that correlations between attention scores and predictions do not constitute biological validation.

Response to Reviewer 2:

This manuscript describes the introduction of the Genotype-to-Phenotype Transformer (G2PT), described by the authors as "a framework for modeling hierarchical information flow among variants, genes, multigenic systems, and phenotypes." The authors used the ratio TG/HDL as a trait for proof of concept of this tool.

This is a potentially interesting computational tool of interest to bioinformaticians, computational genomicists, and biologists.

We thank the reviewer for their overall positive assessment of our study.

Comment 2-1. The rationale for choosing the TG/HDL ratio for this proof of concept analysis is not well justified beyond it being a marker for insulin resistance. Overall the use of a ratio may be problematic (see below). Analyses of TG and HDL separately as individual quantitative traits would be of interest. And an analysis of a dichotomous clinical trait (T2DM or CAD) would also be of great interest.

Reply 2-1. We thank the reviewer for this suggestion. In the revised manuscript, we have expanded our analyses beyond the TG/HDL ratio to include TG and HDL as individual quantitative traits (Fig. 2, reproduced below). These additional analyses demonstrate that G2PT captures predictive signals robustly across each lipid component, not solely through their ratio. Furthermore, to address the reviewer's interest in clinical outcomes, we incorporated an analysis of type 2 diabetes (T2D) as a dichotomous trait of direct clinical relevance. Collectively, these results strengthen the rationale for our chosen phenotype and show that the G2PT framework generalizes effectively across quantitative and binary traits, consistently outperforming advanced PRS and machine learning benchmarks.

Comment 2-2. The approach to mapping SNPs to genes does not incorporate the most advanced approaches. This should be described in more detail.

Reply 2-2. We agree that the choice of SNP-to-gene mapping materially affects both performance and interpretability-indeed, our epistasis simulations suggest that more accurate mappings can improve recovery and localization. In this proof-of-concept work we use a straightforward, modular mapping sufficient to demonstrate the modeling framework, and we have clarified this in the Methods. The architecture is designed to plug-and-play alternative SNP-to-gene maps (e.g., eQTL/colocalization-based assignments, promoter-capture Hi-C). A dedicated follow-up study will systematically compare these alternatives and quantify their impact on attribution and downstream discovery.

Comment 2-3. The example of gene prioritization at the A1/C3/A4/A5 gene locus is not particularly illuminating, as the prioritized genes are already well-known to influence TG and HDL-C levels and the TG/HDL ratio. Can the authors provide an example where G2PT prioritized a gene at a locus that is not already a well-known regulator of TG and HDL metabolism?

Reply 2-3. We thank the reviewer for this suggestion. We have revised the manuscript to de-emphasize the well-established APOA1 locus and instead highlight the less expected "Positive regulation of immunoglobulin production" system (Figure 3a,b, Discussion). Here our model prioritizes the gene TNFSF13 based on specific variants that are not previously associated with TG or HDL (e.g., rs5030405, rs1858406, shown in blue). This finding points to an intriguing, non-canonical link between B-cell regulation and lipid metabolism. While full exploration of this finding is beyond the scope of the present methods paper, this example demonstrates G2PT's ability to identify novel, high-priority candidates in atypical systems.

Comment 2-4. The identification of epistatic interactions is a potentially interesting application of G2PT. However, suppl table 1 shows a very limited number of such interactions with even fewer genes, and most of these are well established biological interactions (such as LPL/apoA5). The TGFB1 and FKBP1A interaction is interesting and should be discussed. What is needed for increasing the number of potential interactions, greater power?

Reply 2-4. We are glad the reviewer appreciates the use of the G2PT model to identify epistatic interactions. We have now discussed a potential mechanism of epistasis between TGFB1 and FKBP1A in the protein dephosphorylation system (Discussion). In addition, we have addressed the reviewer's question about statistical power through extensive epistasis simulations (Fig. 5 and Supplementary Fig. 6), which show that G2PT's detection ability scales strongly with sample size-1,000 samples are insufficient, performance improves at 5,000, and power becomes reliable at 100,000. Realistic simulations (Fig. 5b-d) further demonstrate that under biologically plausible architectures, G2PT can robustly recover specific interactions even within complex genetic backgrounds

Comment 2-5. Furthermore, the use of the TG/HDL ratio for the assessment of epistatic interactions may be problematic. For example, if one SNP affected only TG and the other only HDL-C, it would appear to be an epistatic interaction with regard to the ratio, although the biological epistasis may be limited to non-existent.

Reply 2-5. We have greatly expanded the example phenotypes modeled in our study, Please see our reply 2-1 above.

Response to Reviewer 3:

This manuscript by Lee et al provides a sensible and powerful approach to polygenic score prediction. The model aggregates information from SNPs to genes to systems, using a transformer based architecture, which appears to increase predictive performance, produce interpretable outputs of genes and systems that underlie risk, and identify candidates for epistasis tests.

I think the manuscript is clear and well written, and conducted via state-of-the-art approaches. I don't have any concerns regarding the claims that are made.

We thank the reviewer for their very positive assessment of our study.

Major comments:

Comment 3-1. Specifically, lipid based traits are perhaps the most well-powered and the most biologically coherent; they are also very well-studied biologically and thus overrepresented in the gene ontology. It is unclear whether this approach will work as well for a trait like Schizophrenia for which the underlying pathways are not as well captured in existing ontologies. The authors anticipate this in their limitations section, and I am not expecting them to solve every issue with this, but it would be nice to expand the testing a little bit beyond only this one trait.

Reply 3-1. We appreciate the reviewer's suggestion to expand beyond a single lipid trait. In the revised manuscript, we have included analyses of additional phenotypes, including low-density lipoprotein (LDL) and T2D (Fig. 2). These additions demonstrate the broader applicability of our framework beyond a single trait class.

Comment 3-2. It also seems like the authors have not compared their method to the truly latest PRS methods, such as PRS-CSx and SBayesR. I would suggest adding some of the methods shown to be the best from this recent paper: https://www.nature.com/articles/s41598-025-02903-1

Reply 3-2. We agree these are important comparators. Accordingly, we have extended our comparison to include PRS‑CS (Ge et al., 2019) and SBayesRC (Zheng et al., 2024), following its strong performance demonstrated in recent benchmarking studies (see Figure 2 above). We confirmed that G2PT outperforms advanced PRS methods for all TG/HDL ratio, LDL, and T2D phenotypes.

Comment 3-3. Another major comment regards whether this method could be applied to traits with just GWAS summary statistics, rather than individual level data. This would not enable identification of specific methods underlying an individual, but it could still learn SNP based weights that could be mapped to genes and systems that could help explain risk when the model is applied to individuals (kind of like a pretraining step?)

Reply 3-3. We appreciate this suggestion. While SNP weights from GWAS summary statistics could, in principle, serve as informative priors for attention values, incorporating them would require a sophisticated mathematical formulation that is beyond the scope of this study. Our current framework also relies on individual-level genotype and phenotype data to capture multilevel information flow and individual-specific variation.

Minor comments:

Comment 3-4. Why the need to constrain to a small number of SNPs? Is it just computational cost? If so, what would happen as power increases and more SNPs exceed the thresholds used?

Reply 3-4. Yes, it's about computational cost, but we've now modified the code for improved computational efficiency. First, we refactored the model to use the xFormer memory-efficient attention for the hierarchical graph transformer (Lefaudeux et al., 2022), which also helps full parallelization of training, reducing bottleneck effects. Second, we added a scaling study of the impact of varying SNP count. On 4×A30 GPUs, end-to-end training time for the 5k-SNP setting decreased from 65 hours to 7 GPU-hours (×9). We expect performance can potentially increase if more SNPs are provided to the model based on Fig. 2 (reproduced above). With the optimized implementation, users can raise SNP thresholds as power increases; the expected behavior is improved accuracy up to a plateau, while hierarchical sparsity maintains training tractability and ensures well-regularized results.

Comment 3-5. What type of sample size/power does this method require to work well? If others were to use it, how many SNPs/samples would be needed to obtain good performance?

Reply 3-5. To address this comment, we quantified performance as a function of training size by subsampling the cohort and retraining G2PT with identical architecture and SNP set. New Supplementary Fig. 3 (reproduced below) shows monotonic gains with sample size across three representative phenotypes. We found that stable performance is reached by ~100k samples. These trends hold for continuous traits (TG/HDL, LDL) and more modestly for a binary trait (T2D), consistent with lower per-sample information for case-control settings.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #3

Evidence, reproducibility and clarity

This manuscript by Lee et al provides a sensible and powerful approach to polygenic score prediction. The model aggregates information from SNPs to genes to systems, using a transformer based architecture, which appears to increase predictive performance, produce interpretable outputs of genes and systems that underlie risk, and identify candidates for epistasis tests.

I think the manuscript is clear and well written, and conducted via state-of-the-art approaches. I don't have any concerns regarding the claims that are made.

My two major comments regard a question about how well this will work when compared to other approaches for other traits besides TG:HDL. Specifically, lipid based traits are perhaps the most well-powered and the most biologically coherent; they are also very well-studied biologically and thus overrepresented in the gene ontology. It is unclear whether this approach will work as well for a trait like Schizophrenia for which the underlying pathways are not as well captured in existing ontologies. The authors anticipate this in their limitations section, and I am not expecting them to solve every issue with this, but it would be nice to expand the testing a little bit beyond only this one trait.

Therefore, I would suggest that the authors test a limited number of additional traits that are not lipid based traits, and ideally not metabolic traits, to see how their model behaves. I would pick well-powered GWAS with a lot of associations but from a different phenotypic category

It also seems like the authors have not compared their method to the truly latest PRS methods, such as PRS-CSx and SBayesR. I would suggest adding some of the methods shown to be the best from this recent paper: https://www.nature.com/articles/s41598-025-02903-1

Another major comment regards whether this method could be applied to traits with just GWAS summary statistics, rather than individual level data. This would not enable identification of specific methods underlying an individual, but it could still learn SNP based weights that could be mapped to genes and systems that could help explain risk when the model is applied to individuals (kind of like a pretraining step?)

Other minor comments:

Why the need to constrain to a small number of SNPs? Is it just computational cost? If so, what would happen as power increases and more SNPs exceed the thresholds used?

What type of sample size/power does this method require to work well? If others were to use it, how many SNPs/samples would be needed to obtain good performance?

Will this work just as well for binary diseases? Is this a straightforward extension of the method or does it require more work?

Since I think a lot of geneticists will read it, more intuition as to how attention weights map to parameters geneticists think about would be useful, in particular how the graphics in Fig 3 are made (this may be second nature to ML experts but may not be obvious to statistical geneticists)

The authors claim that G2PT identifies epistatic interactions. Is this true or does it just identify pairs of SNPs that could be subsequently tested for epistasis?

Significance

This study does a great job of marrying the latest (interesting) technologies in AI/ML with a specific problem in statistical genetics. The clarity of presentation and interpretability of the model are strong. The main areas for improvement are to clarify how general this approach is -- will it work for other traits, is it truly better than the latest PRS methods, and what are the specifics of the GWAS it requires (sample size, individual-level data, power, type of trait)

I think the main advance is therefore currently conceptual, but not yet practical, unless more performance comparisons were done.

It seems like the main audience would be geneticists, since I suspect most AI/ML researchers are familiar with this type of approach. If there are fundamental innovations in applying transformers in this specific way to genetics, that would be good to highlight in more depth.

My expertise: statistical genetics and computer science, familiar with DNNs but not a practitioner in them.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #2

Evidence, reproducibility and clarity

This manuscript describes the introduction of the Genotype-to-Phenotype Transformer (G2PT), described by the authors as "a framework for modeling hierarchical information flow among variants, genes, multigenic systems, and phenotypes." The authors used the ratio TG/HDL as a trait for proof of concept of this tool.

Specific comments:

1. The rationale for choosing the TG/HDL ratio for this proof of concept analysis is not well justified beyond it being a marker for insulin resistance. Overall the use of a ratio may be problematic (see below). Analyses of TG and HDL separately as individual quantitative traits would be of interest. And an analysis of a dichotomous clinical trait (T2DM or CAD) would also be of great interest.
2. The approach to mapping SNPs to genes does not incorporate the most advanced approaches. This should be described in more detail.
3. The example of gene prioritization at the A1/C3/A4/A5 gene locus is not particularly illuminating, as the prioritized genes are already well-known to influence TG and HDL-C levels and the TG/HDL ratio. Can the authors provide an example where G2PT prioritized a gene at a locus that is not already a well-known regulator of TG and HDL metabolism?
4. The identification of epistatic interactions is a potentially interesting application of G2PT. However, suppl table 1 shows a very limited number of such interactions with even fewer genes, and most of these are well established biological interactions (such as LPL/apoA5). The TGFB1 and FKBP1A interaction is interesting and should be discussed. What is needed for increasing the number of potential interactions, greater power?
5. Furthermore, the use of the TG/HDL ratio for the assessment of epistatic interactions may be problematic. For example, if one SNP affected only TG and the other only HDL-C, it would appear to be an epistatic interaction with regard to the ratio, although the biological epistasis may be limited to non-existent.

Significance

This is a potentially interesting computational tool of interest to bioinformaticians, computational genomicists, and biologists.

The proof of concept offered here using a single ratio is not sufficient to conclude its potential utility.

My expertise is in genetics and molecular mechanisms of lipid metabolism.

Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

Learn more at Review Commons

Referee #1

Evidence, reproducibility and clarity

The authors introduce G2PT, a hierarchical graph transformer model that integrates genetic variants (SNPs), gene annotations, and multigenic systems (Gene Ontology) to predict and interpret complex traits.

Major Comments:

1. Insufficient Specification of Model Architecture: The description of the "hierarchical graph transformer" lacks technical depth. Key implementation details are missing: how node embeddings are initialized for SNPs, genes, and systems; how graph connectivity is defined at each level (e.g., adjacency matrices used in Equations 5-9, the sparsity); justification for the choice of embedding dimension and number of attention heads, including any sensitivity analysis; and the architecture of the feed-forward neural networks (e.g., number of layers, activation functions, and hidden dimensions).
2. No Simulation Studies to Validate Epistasis Detection: The ground truth epistasis interaction should use the ones that have been manually validated by literature. The central claim of discovering epistatic interactions relies heavily on the model's attention mechanism and downstream statistical filtering. However, no simulation studies are presented to validate that G2PT can reliably detect epistasis when ground-truth interactions are known. Demonstrating robust detection of non-additive interactions under varying genetic architectures and noise levels in simulated genotype-phenotype datasets is essential to substantiate the method's core capability.
3. Lack of Justification for Model Complexity and Missing Ablation Insights: While Supplementary Figure 2 presents ablation studies, the manuscript needs to justify the high computational cost (168 GPU hours using 4×A30 GPUs) of the full model. It remains unclear how much performance gain is specifically due to reverse propagation (Equations 8-9), which is claimed to capture biological context. The benefit of using a full Gene Ontology hierarchy versus a flat system list is not quantified. There is also no comparison between bidirectional versus unidirectional propagation. Overall, the added complexity is not empirically shown to be necessary.
4. Non-Equivalent Benchmarking Against PRS Methods: Figure 2 compares G2PT to polygenic risk score (PRS) methods such as LDpred2 and Lassosum, but G2PT is run only on SNPs pre-filtered by marginal association (p-values between 10⁻⁵ and 10⁻⁸), while the PRS methods use genome-wide SNPs. This introduces a strong bias in G2PT's favor by effectively removing noise. A fair comparison would require: (a) running LDpred2 and Lassosum on the same pre-filtered SNP sets as G2PT, or (b) running G2PT on genome-wide or LD-pruned SNP sets. The reported superior performance of G2PT may be driven primarily by this input filtering, not the model architecture.
5. No Details on Hyperparameter Optimization: Although the manuscript mentions grid search for hyperparameter tuning, it provides no information about which parameters were optimized (e.g., learning rate, dropout rate, weight decay, attention dropout, FFNN dimensions), what search space was explored, or what final values were selected. There is also no assessment of how sensitive the model's performance is to these choices. Better transparency would help facilitate reproducibility
6. Absence of Control for Key Confounders: In interpreting attention scores as reflecting genetic relevance (e.g., the role of the immunoglobulin system), the model includes only age, sex, and genetic principal components as covariates. Important confounders such as BMI, alcohol use, or medication (e.g., statins) have not been controlled for. Since TG/HDL levels are strongly influenced by environment and lifestyle, it is entirely plausible that some high-attention features reflect environmental tagging, not biological causality.
7. Oversimplified Treatment of SNP-to-Gene Mapping: The SNP-to-gene mapping strategy combines cS2G, eQTL, and nearest-gene annotations, but the limitations of this approach are not adequately addressed. The manuscript does not specify how conflicts between methods are resolved or what fraction of SNPs map ambiguously to multiple genes. Supplementary Figure 2 shows model performance degrades when using only nearest-gene mapping, but there is no systematic analysis of how mapping uncertainties propagate through the hierarchy and affect attention or interpretation.
8. Naive Scoring of System Importance: The method used to quantify the biological relevance of systems (i.e., correlating attention scores with predicted phenotype values) risks circular reasoning. Since the model is trained to optimize prediction, systems that contribute strongly to prediction will naturally show high correlation-even if they are not biologically causal. No comparison is made with established gene set enrichment methods applied to GWAS summary statistics. The approach lacks an independent benchmark to validate that the "important" systems are biologically meaningful.
9. No External Validation to Assess Generalizability: All evaluations are performed using cross-validation within the UK Biobank. There is no assessment of generalizability to independent cohorts or diverse ancestries. Given population structure, genotyping platform, and phenotype measurement variability, external validation is essential before claiming the method is suitable for broader use in polygenic risk assessment.
10. Computational Burden and Scalability Are Not Addressed: The paper notes that training the model requires 168 GPU hours on 4×A30 GPUs for just ~5,000 SNPs. However, there is no discussion of whether G2PT can scale to larger SNP sets (e.g., genome-wide imputed data) or more complex biological hierarchies (e.g., Reactome pathways). Without addressing scalability, the model's applicability to real-world, large-scale genomic datasets remains unclear.

Minor:

1. Attention Weights as Mechanistic Insight: The paper equates high attention scores with biological importance, for example in highlighting the immunoglobulin system. There is no causal validation showing that altering the highlighted SNPs, genes, or systems has an actual effect on TG/HDL. Attention weights in transformer models are known to sometimes reflect spurious correlations, especially in high-dimensional settings. The correlation between attention scores and predictions (Supplementary Fig. 3a,b) does not constitute biological evidence. The interpretability claims can be restated without supporting functional or causal validation.

Significance

Novelty

This work presents novelty by introducing the first transformer-based model that integrates the GO hierarchy to enable bidirectional mapping between genotype and phenotype. Additionally, the use of attention mechanisms to screen for epistasis offers a novel and computationally efficient alternative to traditional exhaustive SNP-SNP interaction tests.

Impact

Target Audience

• Specialized: Computational biologists working on interpretable machine learning methods in genomics.
• Broader: Geneticists investigating polygenic traits and drug developers focusing on pathway-level therapeutic targets.

Limitations vs. Contributions

While the work presents a clear conceptual advance by incorporating hierarchical biological priors and attention mechanisms, the technical contribution is somewhat limited by its validation on a single trait and the absence of simulation-based benchmarking. Nevertheless, the framework shows potential if extended to other traits and experimentally validated.

Overall Assessment

Recommendation: Major Revision

Strengths:

• Predictive performance appears strong.
• The use of biological priors enables interpretability at the pathway level.

Major Weaknesses:

• The current validation is limited to a single trait, restricting generalizability.
• The manuscript lacks a complete and clear description of the model architecture.
• No simulations are provided to assess the method's ability to recover known epistatic interactions or pathways.

Reviewer Expertise: Machine learning applications in genomics and genetics.

So basically yes, immigrant effects on voting caries massive implications

TLDR, using pews data, they find the same sort of effect, they have done 101 credibility checks and passed them all

One question is whether the effect varies by region of the US, does NE become relatively less republican with more immigrants

TLDR: We checked for other statistical explanations and didn't find any

existing in the same period of time

Confounder

This note illustrates how conventions in language influence our perception of concepts of ethics. In pointing out that “ethics” is usually a plural noun, it is important to recognize that ethics is a system of thinking or a framework rather than a set of several distinct principles. In regard to words like “data” or “media,” it is evident that language is a product of society that is not bound by its original roots in Latin. Rather, it is important to focus on consistency in a given situation rather than a standard form. In regard to ethics, it is important to focus on understanding rather than simply applying a set of principles. In short, it is important to recognize that ethics is not simply a consideration of principles, but rather a consideration of language.

This is a timeline of how a single tweet on a social media platform can quickly go viral. Justine Sacco’s tweet started off reaching a very small audience, but once it was shared on a larger media platform, it became decoupled from its original context and quickly spread. The additional factor that increased public speculation and participation was the unavailability of Justine Sacco during the flight of her plane, which turned the event into a real-time spectacle. The popular hashtag on the trending page illustrates the way online platforms promote a sense of collective surveillance and judgment where viewers are not only responding to the content but are also responding to the responses of other viewers.

CHRONOLOGICAL AGE DOES NOT COMPLETELY CAPTURE A PERSON'S AGE

SOCIAL AGE

it is becoming less relevant in the 21st century

Mary Shelley wrote it as a teenager after a ghost story challenge during a stormy summer in Switzerland, and it's considered the first true work of science fiction, born from a nightmare and inspired by real scientific debates about reanimation. Also, the name Frankenstein belongs to the creator, Victor, not the monster, a common misconception.

there are so many families who are barely above the poverty line and one crisis way from being in poverty just seeing the numbers made me realize there are so many families who are in this situation

by the economist Mollie Orshanky

This is a very romanticized version of the war, kind-of showing how people can sign up for the idealized rendition of something.

this is saying that once you reach this place, you will have eternal peace and there will be nothing bad or evil that can reach you.

this poem tends to use more positive language and talks more about leaving a legacy behind

Characters can people or places so long as they significantly affect the other character's choices. And there are several types of characters.

Allegory uses characters to represent archetypes or deeper themes of society, or humanity etc

I should make this list the basis of my analysis

Analysis involves your own original thoughts, opining on the significance of various themes

Take all the info you have gathered and from that data, create a thesis statement

Create a list of 5-10 words that convey the main idea of each paragraph

Use annotations to keep track of insights and data

Keep the summary as small as possible

Summary is the who what when where and how of a story.

Prob predictable

VS speed typist Albert Tangora typing on a typewriter on table..<br /> by [[Getty Images]]<br /> accessed on 2026-01-16T12:03:59

It's not just the speed at which, he's typing, but get a load of how fast he manages to change a sheet of paper!