GO curation workflow

[cthoyt]

Here's a rough sketch of what a curation workflow for a GO term might look like

Query

1. Get GO term as input
2. Get all genes associated with a GO term and children of GO term
3. Get all statements (that have text evidence) from a union of several queries
   • Statements between genes
   • Pairs of statements mediated by an unassociated entity
   • Optional - pairs of statements that both have associated genes as subjects but a shared unassociated entity as the downstream target
   • Optional - pairs of statements that both have associated genes as targets but a shared unassociated entity as the upstream controller

Curation

1. Convert statements to directed graph
2. Sort edges by importance taking into account potentially several factors
   • Evidence counts (depending on if you want to prioritize curating canonical or potentially obscure biology)
   • Belief score
   • Edge centrality / betweenness centrality
3. Filter and re-sort by previous curation
   • If a statement has conflicting curations (e.g., both more 1 positive and more 1 negative curation), then it should be curated a third time. These should be put all the way at the top
   • If a statement has been curated positive or negative 2 or more times and doesn't have conflicts, it doesn't need to be curated again.