Run a generative AI chatbot on Jetson Orin Nano Super Developer Kit. This chatbot features Ollama with Open WebUI, a widely used, open-source, chatbot server interface that connects to locally running LLMs.

npm is a couple of things. First and foremost, it is an online repository for the publishing of open-source Node.js projects. Second, it is a CLI tool that aids you install those packages and manage their versions and dependencies.

our model solves 28.8% of the problems, while GPT-3 solves 0% and GPT-J solves 11.4%

translates whole-programs from one high-level programming language to another

symbolicexecution (symexec)

theoretical framework is needed to represent the complex interactions that drive these dynamics

successful establishment of the organism into the community

Omi: Bridging the Bioinformatics Expertise Gap with a Natural Language-Driven Metagenomics Co-pilot

Thankfully, the development of AI technologies, especially Large Language Models (LLMs) [8, 9, 10] with strong reasoning, adequate knowledge reserve and excellent coding capabilities [11], is reshaping the paradigms and precepts of how people leverage bioinformatics data.

Domain specialization as the key to make large language models disruptive

Domain specification techniques are key to make large language models disruptive in many applications

many hurdles, caused by the heterogeneity of domain data, the sophistication of domain knowledge, the uniqueness of domain objectives, and the diversity of the constraints

we present a comprehensive survey on domain specification techniques for large language models, an emerging direction critical for large language model applications

LLMs, significantly outperforming smaller models in understanding and generating human-like text,have emerged as a promising AI research trend

efficient literature analysis, novel hypothesis generation, and complex data interpretation

domainspecialization of Large Language Models (LLMs) is defined as the process of customizing general-purpose LLMs accordingto specific domain contextual data, augmented by domain-specific knowledge, optimized by the domain’s objective, andregulated by domain-specific constraints

Parameter-efficient fine-tuning (PEFT)

additive adapters applied to frozen model weights

suboptimal model quality due to restrictive assumptions

sketching, a popular data compression technique, can serve as an efficient adaptation strategy for LLMs while avoiding low-rank assumptions

compresses LLM weights into compact fine-tunable sketches

faster and more memory-efficient training and inference

evaluations with Llama-1/2/3 models

diverse tasks including math problem-solving, common sense reasoning, and instruction following

labor-intensive workflows

SpatialAgent, a fully autonomous AI agent

SpatialAgent integrates large language models with dynamic tool execution and adaptive reasoning

combining autonomy with human collaboration

complex high dimensional data

computational methods that are quite fragmented

extensive human intervention

users have to reason over multiple analysis approaches

require deep biological knowledge

from experimental design to multimodal analysis and data-driven hypothesis generation

predefined workflows and rigid models, SpatialAgent employs adaptive reasoning and dynamic tool integration, allowing it to adjust to new datasets, tissue types, and biological questions

supports human-in-the-loop interactions

tasks such as gene panel design, cell and tissue annotation, and pattern inference in cell-cell communication and pathway analysis

Key modules. The action module (left) executes tasks such as retrieving reference datasets, converting gene names, verifying ligand–receptor interactions using existing databases, processing data with established software packages (e.g., numpy) or generating and executing custom code, while reasoning over and aggregating information from multiple sources

SpatialAgent consists of three key modules (Fig. 1b): memory, planning, and action.

Wratten, L., Wilm, A. & Göke, J. Reproducible, scalable, and shareable analysis pipelines with bioinformatics workflow managers. Nat Methods 1–8 (2021) doi:10.1038/s41592-021-01254-9.

barrier is overcome by workflows, which implement popular analysis strategies

accessible to those who have limited training in computation

the ability to analyse data beyond predefined strategies they implement continues to be largely limited to ‘master chefs’

microbiologists who wanted more freedom in research questions they could ask of their data

interactive refinement

interactive visualisation and editing software

more than 100 interoperable programmes

anvi’o empowers its users to navigate through ‘omics data without imposing rigid workflows.

Nano (8B)

open family of heterogeneous reasoning models that deliver exceptional reasoning capabilities, inference efficiency, and an open license for enterprise use.

performs competitively with state-of-the-art reasoning models such as DeepSeek-R1 while offering superior inference throughput and memory efficiency

using neural architecture search from Llama 3 models for accelerated inference, knowledge distillation, and continued pretraining, followed by a reasoning-focused post-training stage

support a dynamic reasoning toggle, allowing users to switch between standard chat and reasoning modes during inference

early tools such as NPLinker190, GraphOmics200 and anvi’o201 are starting to overcome some of these challenges

Another bottleneck is getting the data in the appropriate format so it can be used by AI algorithms

users still needing considerable expertise to interpret the results.

inter-dependencies of the data types and the various data formats that need to ‘talk’ to each other.

anvi’o75 was used to profile and visualize the different Turicibacter strain DNA sequences to locate putative bile salt hydrolase and 7α-HSDH homologs in contig groups, generate variability profiles, and measure gene coverage and detection statistics.

Yet, taxonomic insights offer limited utility to understand functional drivers of biological systems, a pinnacle desire that brings together many corners of microbiology

Understanding of microbes in their environmental context required genomes

ability to reconstruct microbial genomes directly from the environment

can provide clues regarding metabolic requirements of little-known organisms and guide their directed isolation

careful interpretations of variants observed in metagenomes are becoming more accessible through integrative ‘omics platforms223

computational resources and large datasets required, however, limit their applicability in biological contexts

scaling up foundation and task-specific models

introduce Lyra, a subquadratic architecture for sequence modeling, grounded in the biological framework of epistasis for understanding sequence-to-function relationships

plasmid binning, that is identifying plasmids in sequenced bacterial isolates

we propose the use of a pangenome graph, built from assembly graphs produced by assembling short reads of the same sample with different assemblers

highlights similarities between contigs from different assemblies while retaining information on contigs that appear only in one of the input assemblies

Assembly graphs produced by different tools from the same data may differ significantly, posing a challenge to tools for downstream processing tasks

Constructing a pangenome graph, however, is still a time-consuming and expensive process.

augment a pangenome graph using unassembled reads

does not require to align them and genotype the new individuals

present the first assembly-free and mapping-free approach for augmenting an existing pangenome graph using unassembled long reads from an individual not already present in the pangenome.

agentic technology uses tool calling on the backend to obtain up-to-date information, optimize workflows and create subtasks autonomously to achieve complex goals.

autonomous agent learns to adapt to user expectations over time.

ability to store past interactions in memory and plan future actions encourages a personalized experience and comprehensive responses

agent can update its knowledge base and perform agentic reasoning

each step of the way, the agent reassesses its plan of action and self-corrects, allowing for informed decision-making.

scaffold information generated by Bambus 2 allows us to integrate multiple sources of information and obtain more accurate annotations of the resulting assembly

Unlike MetAMOS, SmashCommunity only supports a small set of assembly and analysis tools

simply links together the individual analysis tools

provide additional functionality made possible by the integration of different analyses

compare its performance to other software tools

'Assembly mode', which requires larger amounts of RAM and starts from raw read data

We intended to encourage users to tailor MetAMOS to the biological questions they want to answer, not the inverse

customize their own pipelines by combining the modules they deem necessary

Ruffus [29]) to track inputs/outputs/states and checkpoint while running through computationally intensive analyses.

INSTALL script. This will automatically configure the pipeline to run within the user's environment and also fetch all required data

initPipeline is mainly involved with creating a project environment, and describing input files

runPipeline takes a project directory as the input and will initiate execution of the entire MetAMOS pipeline

MetAMOS pipeline ends by generating an interactive, HTML summary

assembly statistics and estimated abundance information

commercial 16S bioinformatic pipeline from 1928 Diagnostics (1928-16S) was evaluated and compared with the open-sourced gms_16S pipeline that is based on the EMU classification tool (GMS-16S).

RapidONT, a workflow designed for cost-effective and accessible WGS-based pathogen analysis

routine clinical adoption of WGS is hindered by factors such as high costs, technical complexity, and the requirement for bioinformatics expertise for data analysis

user-friendly web-based platform Pathogenwatch, which facilitates species identification, molecular typing, and antimicrobial resistance (AMR) prediction

choice of the right algorithm for a given dataset has become difficult due to numerous comparative reports on these different assemblers [88, 89]

most widely used assemblers are MegaHit, metaSPAdes, RayMeta and IDBA-UD

The aim of this work is to review the most important workflows for 16S rRNA sequencing and shotgun and long-read metagenomics

assembly, binning, annotation and visualization

best-practice protocols

major advantage of De Bruijn graphs is that assembled reads contain fewer errors and errors can be easily corrected prior to assembly

A Modular, Open-Source whole genome assembler.

human mtDNA is an extranuclear molecule of ∼16.5 kilobases

mtDNA is an informative matrilineal uniparental marker that can be used to trace the ancestry of an individual

scanned all the assembled contigs from each sample for human mtDNA by running homology search (i.e., BLASTn)

Positive cases were derived from stool, oral, and skin samples

it is now considered mandatory to remove human DNA (or RNA) sequencing reads before depositing metagenomes in public repositories

Human DNA could be considered personal identifying information

there is not a consensus on which version of the human reference genome to use for human DNA decontamination

studies that reported exclusion of only reads where both paired-end reads are mapped still detected mtDNA

small, circular nature of the mitochondrial genome allows reads to span the start and end positions, leading to incomplete exclusion of mtDNA

nuclear DNA is linear and much larger, so this approach effectively removes most nuclear DNA reads, leaving only minimal traces

exclude more off-target reads

using single-end mapping

extensive validation of the available pipelines is still required.

MADRe, a modular and scalable pipeline for long-read strain-level metagenomic classification, enhanced with Metagenome Assembly-Driven Database Reduction.

contig-to-reference mapping reassignment based on an expectation-maximization algorithm for database reduction,

mapping-based tools such as MetaMaps [24], PathoScope2 [25], EMU [26] and MORA [27], which rely on read alignments and reassignment algorithms, offer higher precision at a greater computational cost.

Kraken2, perform well at the species level

The implementation of the EM algorithm in MADRe is inspired by PathoScope2 [46] and EMU [26].

range of metagenomic classification tools have been developed, which can be broadly categorized into marker-based, DNA-to-protein and DNA-to-DNA approaches, as described in [4].

K-mer-based tools such as Kraken2 [14], KrakenUniq [15], Bracken [16], Centrifuge [17], CLARK/CLARKS [18, 19], Ganon [20, 21], Taxor [22], and Sylph [23] are known for their speed and scalability to large databases, but often trade precision for speed

MADRe achieves high precision and strain-level resolution while maintaining lower memory usage and runtime compared to existing tools

assembly tools remain prone to large-scale errors caused by repeats in the genome, leading to inaccurate detection of AMR gene content

we present Amira, a tool to detect AMR genes directly from unassembled long-read sequencing data

the fact that multiple consecutive genes lie within a single read to construct gene-space de Bruijn graphs where the k-mer alphabet is the set of genes in the pan-genome of the species under study

reads corresponding to different copies of AMR genes can be effectively separated based on the genomic context of the AMR genes, and used to infer the nucleotide sequence of each copy

compare the number of fully (>90%) present genes with good read support by Amira and Flye with AMRFinderPlus

quantifying the improvement in recall when handling heterogeneous data.

We present Autocycler, a command-line tool for generating accurate bacterial genome assemblies by combining multiple alternative long-read assemblies of the same genome

Autocycler builds a compacted De Bruijn graph from the input assemblies, clusters and filters contigs, trims overlaps and resolves consensus sequences by selecting the most common variant at each locus

“module scripts” (or “modules” for short), which are Nextflow scripts that can be “included” by other scripts

help you organize a large pipeline into multiple smaller files and take advantage of modules created by others

To migrate this code to DSL2, you need to move all of your channel logic throughout the script into a workflow definition

initial goal of tinybio was to remove the barrier to entry for running bioinformatics packages

Scientists spend a significant proportion of their time transforming and structuring data for analysis

driving the development of community-centric tools on Seqera.io, empowering scientists worldwide to leverage modern software capabilities on demand

removing barriers to entry to bioinformatics

steep learning curve that prevents newcomers from getting started fast

meet scientists at every stage of their work

Suggesting

Generating Nextflow code

Asking bioinformatics questions

contextually relevant answers

Beyond just a chat interface

ability to test their code in the interface

Programmed with a deep understanding of Nextflow, common bioinformatics tools, and the overarching scientific community.

extensive testing with scientists

able to identify the root cause of errors, help troubleshoot, and suggest edits

has deep knowledge of the errors

ability to pair with bioinformatics test data and generate local test scripts

generate and run unit tests.

not only give you the initial conversion, but also run the stages of the code that it generates with sample data and iteratively correct any code that yields runtime errors

convert a pipeline from Bash/CWL/WDL to Nextflow

AI can be a powerful tool for helping scientists dig into results and more quickly identify interesting patterns

key to figure out how we can get the right context on your pipeline results

Seqera AI – a bioinformatics agent purpose-built for the scientific lifecycle

native integration with MultiQC where you can enable automatic, in-line analysis of MultiQC reports

fully extensible endpoint in Seqera AI, so that any bioinformatics tool can build their own AI integration.

threatens to compound this problem owing to the ease with which massive volumes of synthetic data can be generated

importance of improved educational programs aimed at biologists and life scientists that emphasize best practices in data engineering

increased theoretical and empirical research on data provenance, error propagation, and on understanding the impact of errors on analytic pipelines

we focus specifically on concerns that lie at the interface of biological data and computational inference with the goal of inspiring increased research and educational activities in this space

how to best benefit from recent advances in AI and how to generate, format and disseminate data to enable future breakthroughs in AI-guided drug discovery

it supports both short and long reads

When given well-crafted instructions, these chatbots hold the potential to significantly augment bioinformatics education and research

Crafting effective prompts can be challenging

role prompting that assigns a role to the chatbot, few-shot prompting that provides relevant examples, and chatbot self-reflection that improves responses based on task feedbacks

domain-specific knowledge.

In addition, varying study designs will require project-specific statistical analyses.

Hecatomb’s design philosophy recognizes that there are no “perfect” databases or search algorithms

Instead, Hecatomb relies on providing a compiled and rich set of data for search result evaluation

Hecatomb and Conda handle the installation of all dependencies

use of isolated Conda environments for Hecatomb minimizes package version conflicts, minimizes overhead when rebuilding environments for updated dependencies, and allows maintenance and customization of different Hecatomb versions.

While Hecatomb is a Snakemake pipeline, it uses the Snaketool command line interface to make running the pipeline as simple as possible [95]. Snaketool populates required file paths and configuration files, allowing Hecatomb to be configured and run with a simple command

An opt-in feature for now, strict syntax enables consistent behavior between the Nextflow CLI and language server, and enables numerous new features

more actionable error messages

output on the terminal highlighting exactly where the problem lies

This new specification enables more specific error reporting, ensures more consistent code, and will allow the Nextflow language to evolve independently of Groovy.

strict syntax will eventually become the only way to write Nextflow code, and new language features will be implemented only in the strict syntax

prepare for the strict syntax

assignments are allowed only as statements:

use higher-order functions, such as the each method, instead:

use if-else statements

environment variables

Use a multi-line string

Any Groovy code can be moved into the lib directory, which supports the full Groovy language.

For Groovy code that is complicated or if it depends on third-party libraries, it may be better to create a plugin

local executor is very useful for workflow development and testing purposes

Nextflow provides an abstraction between the workflow’s functional logic and the underlying execution system (or runtime)

its cost-effectiveness and lower data requirements compared to metagenomic whole-genome sequencing (WGS)

interactive analysis applications

namely Jupyter, RStudio, VS Code, and Xpra).

You should always be suspicious of your metabolic reconstructions, and particularly when you are using short reads where you have partial matches.