74 Matching Annotations
  1. Feb 2020
    1. The parametervector for a classcis~θc={θc1,θc2,...,θcn}

      needed for CNB formula

    2. However, the class probabilities tend to beoverpowered by the combination of word probabilities,so we use a uniform prior estimate for simplicity

      what is the uniform prior estimate for the class?

  2. Dec 2019
    1. ages 2–10 y
    2. average stature
    3. height velocity
    4. 0.8 cm/mo (10 cm/y)
    5. 1 y
    6. median height velocity
    7. 1 y of life
    8. infants
    9. ≈1.7 cm/mo (20 cm/y)
    10. ≈3.7 cm/mo (44 cm/y)
    11. shortly after birth
    12. average stature
    13. median height velocity
    14. average stature
    15. puberty
    16. infancy
    17. height velocity
    18. average stature
    19. height velocity
    20. average stature
    21. height
    22. average stature
    23. height
    24. 48.5 ± 0.1; n = 595
    25. 47.4 ± 0.2 cm; n = 242

      value, sd, n

    26. birth length
    27. average stature
    28. mean birth length
    29. average stature
    30. Height
    31. average-stature
    32. median weight-for-age
    33. average stature
    34. 5th
  3. Nov 2019
    1. Ref. [44], and for BioNLP, in Ref. [45].

      useful for me

    2. Add handling of negation and hedging


    3. (or the entire document)


    4. affects recall


    5. ActualRelatedNot RelatedPredictedRelated2010Not Related2731Precision = 67%Recall = 43%

      i am not sure that this is an improvement if you are losing 57% of your recall! depends of what you value more, novel relationship discover or accuracy

    6. define a set of selected features for each one.

      what features? PO? symantics? entity patterns?

    7. OMIM MorbidMap

      not very surprising due to the size of omim and the inevitable gaps in the data

    8. The simple NER approach that we applied to medical records, leveraging Uruguayan terminology dictionaries from SNOMED CT, produced the results listed in Table 3 for the domains of disease and finding entities. A physician manually evaluated the expected vs. found entities for each medical record, considering an entity as correctly identified when its recognized name was exactly or nearly exactly as expected. We obtained a precision of 94% when counting entity repetitions (more than one mention), and 87% leaving repetitions apart

      not mentioning negation which is important

    9. Fig. 5. Structure of the mini ontology: A graph with nodes representing entities and edges representing relations between them.

      i note there is no connection between pubmed and phenotype which suggests that the genotype-phenotype is omim based. OMIM has quite shallow phenotyping as i am aware

    10. hipertensión arterial (English arterial hypertension), the dictionary-based approach correctly recognizes hipertensión arterial disease in Spanish, but the CoreNLP node recognized both hipertensión (English hypertension) and arterial only individually.

      sounds like a language incompatibility as arterial hypertension is not a term used in english medicine

    11. Since an actual set of medical records in Spanish was not available for research, we manually transcribed 109 clinical notes with physician observations, from actual patient cases, used for medical education.

      this type of case will be succinct and cherry picked for educational purposes - will need real life validation

    12. The knowledge base can be queried in two ways: 1) starting from the medical record, and leading to related entities (like genes); or 2) starting from genes of interest (previously obtained from patients genome or exome analysis), and leading to related diseases and substances

      This all sounds very useful as an approach to a new case clincally

    13. Mini Ontology, is permanently updated from a corpus that contains novel articles, on a daily basis


    14. {relation_type, entitity_1, entity_2}.

      must predefine the relations

    15. Wu et al. [22] compare alternatives based on word embeddings to improve NER results in BioNLP, against existent proposals based on CRF, MaxEnt, and SVM. Chiu et al. [23] devise guidelines for good word2vec based embeddings, both CBOW and skip-gram, working on PubMed and the PMC corpus. For auxiliary tasks, these authors use GeniaSS as a sentence splitter and NLTK [24] for word tokenizing

      The challenges of word embedding and text processing in a specialised domain

    16. GeniaSS as a sentence splitter

      Will be useful to add to my pipeline

    17. Genia Tagger [13] has been frequently used both for part of speech tagging and named entity recognition (NER) in the BioNLP domain. For Spanish medical documents, Genia Tagger has been used in conjunction with Freeling [14] for entity recognition and automatic annotation [15]

      NER annotations

    18. HGVS

      genomic variant ontolgy

    19. OMIM

      genotype phenotype ontology

    20. SNOMED CT,

      medical language ontology - very large

    21. The goal of this work is to provide tools for the medical geneticist that optimize his/her access to the latest research pertaining to a specific patient (or to specific genomic information)

      SUggesting the reverse and forward genetics genotype - phenotype (need high prob variants) phenotype - genotype (need good phenotyping) approaches

    22. This makes attempts to become properly acquainted with the latest findings that could be relevant to a specific patient particularly challenging


    23. Are there known substance/drug interactions?;

      less relevant in clinical genetics

    24. Is this variant pathogenic?; With which phenotypes/diseases is this variant associated?;

      All relevant

    25. The literature requires reviewing in such a way that will allow the gathering of the latest findings

      Case reports sequencing cohorts case series reviews and meta-analysis contemporaneous literature is key for rare conditions

    26. A patient genome can be sequenced in a few hours or even minutes [1],

      Not much use without the analysis that supports any interpretation

    27. PubMed abstracts

      Key advantage we should have with full text analysis

  4. Jun 2019
  5. May 2019
  6. Apr 2019
    1. screening, surveillance, and interventional measures,

      will need to consider locally available options for all of these

    2. excess toxicities withparticular cytotoxic therapies

      indication for genetic testing