Disease similarity from MEDLINE topic cooccurrence

We have adopted a literature mining scheme for term relations based on cooccurrence of MeSH topics in MEDLINE. Of DO Slim diseases, 133 out of 137 have a MeSH cross-reference. For each disease, we identified all studies assigned that topic and then computed pairwise cooccurrences (notebook, tsv).

Since disease topics seemed averse to cooccurrence, we did not enforce the major topic filter. Even so, diseases cooccurred less than would be expected by chance. This makes sense given that papers often focus on a single disease only but means that our p-values are likely conservative.