Data inconsistency - Compounds

[rafael-alcantara]

Currently, compounds shown in the list of filters (search results) and those shown in the small molecules tab don't match.

The tab seems to be fine, including:

• bioactive compounds (ChEMBL entries from our mega-map)
• drugs (DrugBank IDs from UniProt, converted into ChEBI IDs whenever possible)
• activators/inhibitors (from UniProt, through rough text-mining and converted into ChEBI IDs whenever possible)
• cofactors (from IntEnz, as ChEBI IDs)

However, the compound filter included this:

• ChEMBL entries with a name, from our mega-map
• ChEBI IDs related to the UniProt accessions via CATALYTIC ACTIVITY or COFACTOR lines. These are text-mined by ChEBI during the xrefs import, which results in a fuzzy list (ex. "zinc ion" can be resolved by ChEBI as "zinc", "zinc ion" and "ion", the three of them).

I suggest the following:

• small molecules tab: leave it as is.
• compound filters in search results (taken from mega-map):
  • leave the ChEMBL entries.
  • remove the current ChEBI entries and replace them with:
  • add cofactors from IntEnz (more accurate than text-mining of COFACTOR line)
  • add reaction participants from Rhea (more accurate than text-mining of CATALYTIC ACTIVITY line). Of course, these won't show in the small molecules tab but in the reactions/pathways one.
  • add drugs from UniProt (cross-references to DrugBank)
  • add activators/inhibitors from UniProt (from "Enzyme regulation" comments, text-mining)

All these additions should be done to the mega-map (i.e. use IntEnz, Rhea and extract more from UniProt when building it).

[rafael-alcantara]

mega-mapper has been modified so that the database is populated according to the suggestion above. Furthermore, there is a new table in the database for quick retrieval of compound cross-references.

ep-util is used now for text parsing (inhibitors and activators). Any improvements in regular expressions for that matter will have to be done there.

ep-domain has been modified (XML schema) in order to have roles for molecules, which allows classification of compounds retrieved from the mega-map, and hence the use of the database as the only source of small molecule data, be it for filters or for the small molecules tab (the purpose of this ticket). Other modules which use the domain model, like chebi-adapter, have been updated accordingly.