================================================================================
Rhea NLP
================================================================================

This directory contains data to support the development of natural language
processing (NLP) methods to mine biochemical reactions from text for Rhea and
UniProt.

List of files:

- EnzChemRED.tar.gz

  Content: The EnzChemRED corpus of 1,210 PubMed abstracts with annotations of
  gene/protein and chemical mentions using UniProtKB and ChEBI respectively,
  chemical conversions - relations that link pairs of chemical mentions - and
  the enzymes that catalyze those conversions, when available.
  The development of EnzChemRED is described in Lai, P.-T., et al. (2024).

  File format: BioC - see: 
  Comeau, D.C., et al. (2013) BioC: a minimalist approach to interoperability
  for biomedical text processing. Database (Oxford), 2013, bat064.
  DOI:10.1093/database/bat064; PMID:24048470; PMCID:PMC3889917

- BioREx_EnzChemRED_PubMed.tsv

  Content: Raw results of scanning PubMed abstracts up to December 2023 using a
  BioREx model fine-tuned using EnzChemRED, as described in Lai, P.-T., et al.
  (2024).

  File format: TSV with fields:
   1. ChEBI ID 1
   2. Chemical name 1
   3. ChEBI ID 2
   4. Chemical name 2
   5. Relation type
      Values: Indirect_conversion, Conversion, Non_conversion
   6. BioREx score
      Values: Number ranging from 0 to 1
   7. PMID
   8. Sentence
      Sentence from PMID containing the relation.

- BioREx_EnzChemRED_PubMed_normalized.tsv 

  Content: Processed results of scanning PubMed abstracts up to December 2023
  using a BioREx model fine-tuned using EnzChemRED, as described in Lai, P.-T.,
  et al. (2024).
  ChEBI compounds were first normalized based on the pH7.3 relationships used in
  Rhea, then by the first two layers of their InChI keys for compounds with
  InChI. Pairs with identical ChEBI ID 1 and 2 and pairs where at least one
  ChEBI ID is a participant in >= 100 Rhea reactions were removed from the
  dataset.

  File format: TSV with fields:
   1. ChEBI ID 1
   2. ChEBI name 1
   3. ChEBI ID 2
   4. ChEBI name 2
   5. Relation type
      Values: Indirect_conversion, Conversion, Non_conversion
      For a ChEBI pair found with different relation types in different PMIDs,
      the relation type for the pair is chosen according to the order in which
      the values are listed above.
   6. Max BioREx score
      Values: Number ranging from 0 to 1
      Maximum BioREx score of all relations that include this ChEBI pair.
   7. PMIDs
      Number of PMIDs in which the ChEBI pair appears in the PubMed dataset.
   8. Rhea
      Values: yes, no
      Indicates whether the ChEBI pair is found in Rhea reactions.
   9. Atom conservation (OPTIONAL)
      Percent atom conservation of the ChEBI pair.
  10. Min DRFP distance (OPTIONAL)
      Values: Number ranging from 0 to 2
      Minimum cosine distance between the DRFP of the ChEBI pair to the DRFPs of
      ChEBI pairs from Rhea reactions.