Textpresso Central

General considerations: Specification of data models, markup languages, and flow now is important.

Searching and Category/Ontology Development

• Control panel: loading papers from existing corpora into a viewer, incorporation of PubMed queries; search results will be used to import full text from PMC or journal site

• Searches - existing corpora, list of external identifiers, combination of both
  • External identifiers - which ones? PMIDs, doi's, MOD paper IDs, others?
    • Use case: Perform a PubMed search and then port the resulting IDs to a Textpresso search
  • Paper Exclusion List - user supplied, from external data file, e.g. Gene Ontology Annotation File (gaf)
    • Use cases: TAIR CCC black list and filtering based upon papers already annotated with Component term in gaf file

• Categories and Keywords
  • Organization of categories by task? e.g. GO curation, Phenotype curation, Expression patterns, etc.
  • Create and display category metadata - source, possible use, version, last updated
  • Restrict search to a subset of a category
    • Use case: FlyBase CCC search using only gene names that start with CG
  • How quickly could new searches be performed with modified categories

• Search Filters
  • Bibliographic filters - year, journal, paper type, etc.
  • Data Type Flagging - NLP results - data models and storage
    • Development of NLP toolbox: pattern matching, statistics, svm, hmm, crf
    • Index of all NLP results for faster querying
    • Display most current precision and recall statistics so users can assess the accuracy of an NLP tool
  • Data Type Flagging - author or curator flags
    • Information stored in postgres on tazendra
  • Textpresso search score cut-off - view scores (mean, range) for curatable papers (this would depend on search criteria)
    • • Use case: search all SVM predicted negatives and return those papers with a score >n
  • Curation Status - integration with curation databases - which ones?
  • View previously made annotations - source? tie to a sentence where possible?
    • Robust back-end infrastructure with internal Textpresso database holding all annotations
      • Adapt data models and tables from postgres curation database on tazendra?

• Textpresso Category and Ontology viewer and editor
  • Stand-alone or interface with curation or both
  • Incorporate statistical analyses (word frequency in positive vs negative sentences, how often is a term the only one from a given category, etc.)

Viewing Search Results

• Viewer: selecting terms, importing them into OA, prepopulating entries of forms; display results from NLP tools; initiate new NLP analyses (pattern matching, statistical, machine learning)

This will require a uniform representation of all machine learning results w.r.t. papers in Textpresso. Annotation markup language comes to my mind.

• Viewing options
  • Sort by score, year, journal (like PubMed)
  • See search results within the context of the paper
    • Paper viewer
      • See existing annotations, if tied to a sentence
      • Additional mark-up options? e.g. alleles, reagents, genes

Annotating and Curating

We would need to develop a markup language (XML) for data flows. This should be a generic as possible.

• OA and its interaction with TC
  • Curators would like to be able to view the search results while curating and make annotations from the true positive sentences.

• Flag the paper and/or sentence(s) as curatable, relevant but not curatable (sentence only), false positive (break down further?), unannotated - basically TP, FP, FN, TN
  • Have ability to annotate NLP results in bulk

• Click on a sentence and, depending upon the curation needs, the curation tool is pre-populated with relevant entities.
  • Would need to know species (usually OK for elegans, could be trickier for mammalian species)

• Click on words to add to category

• Current Textpresso-based curation forms:
  • CCC (Cellular Component Curation) form
    • Pros:
      • sentences are seen on the same page as annotations
      • form pre-populates curation fields with protein names, category terms, and suggested annotations
      • easy to mark sentences if not curatable
    • Cons:
      • duplicating or making multiple annotations is cumbersome
      • don't see term info for proteins or GO terms
      • don't see additional annotations for proteins mentioned in sentences
  • the interaction configuration of the OA
  • Any others? Ask other WB curators.

• Add curator comments to a paper, sentence, term

• Output of curation
  • to Textpresso database
  • to MOD or other project database (e.g., BioGRID)
  • downloadable file - what formats?

Data Models and Flow

• Integrate Textpresso categories (TCAT), NLP results and curator annotation (CA) into one big data class

 Model needs following elements; not all elements are populated at all times
 - term (TCAT: lexicon entry; NLP: term, sentence identified in paper if applicable; CA: term manually annotated)
 - annotation (TCAT: category term with possible attributes; NLP: machine-learningID or describing term; CA: manual annotation)
 - paper location: PaperID, SentenceID, PosID
 - allowed lexical variations (plural, tenses)
 - ownership (who can change entry)
 - what else?
 - timestamp
 - version
 - source
 - comment
 - possible use

• Data flow / Transaction model

 - does one big model for all exchanges between all module work?
 ...

Action Items 2011-11-29

• Develop a controlled vocabulary for all items that needs to be query-able:
  • Data type
  • Curation status
  • Possible use
  • Source
  • Type of Annotation
  • Lexical variation

• Make a first version of the annotation data model

• Think about how to track changes in tokenization of papers in annotation database
  • When papers get reformatted (sentence identification improve) how to port old annotations

• Manifestation of Data Model in Textpresso database

• • Import current SVM results
  • Import current HMM results
  • Import current CCC results
  • Import current gene-gene interaction results
  • Import current Molecules results
  • How will category markup go into Textpresso database?

• Design (and later implement) first version of Textpresso Curator interface