implement filtering/ranking based on PFOCR co-occurrence

[andrewsu]

Per 2021-01-19 meeting, tagging @kevinxin90 and @ariutta

[ariutta]

@kevinxin90, do you already have a preference for how to do this co-occurrence, e.g., weights, co-occurrence of pairs or co-occurrence matrices?

[ariutta]

@andrewsu, you asked for a list of the types of data we can get or have gotten from PFOCR, so I'm adding that here.

Have gotten:

• genes
• amino acids
• metabolites / small molecules
• drugs
• diseases (the name of the disease is in the figure)
• bio-terms (e.g., cellular components)

Could get (or in progress or exploratory stage):

• phosphorylation and similar types of state data
• drug, disease and bio-term extraction could be enhanced
• maybe things like biologics (as opposed to conventional drugs)
• could work more with the figure titles, captions and the mentions of the figure in the body of the paper, e.g., "In Fig. 5, we illustrate ..."

Could possibly get:

• localization
• reactions
• cascades
• cycles
• co-factors
• interactions (some cases would be difficult)

Note that for diseases, Alex also did an additional analysis:

Gene Ontology and disease annotations for pathway figures were determined by performing enrichment analysis on the sets of figure-extracted genes against ontology-associated gene sets.

[ariutta]

Closing in favor of new issue: biothings/biothings_explorer#451