We chose three to five common diseases from each of the 13 DO upper-level categories used in our common-disease network (CDN; see below) for a total of 41 diseases. We used a Perl script to choose diseases at random from among all diseases in the categories. We examined the diseases manually by assessing each HPO term mentioned at least once in any abstract describing the disease in question (thus, we evaluated substantially more HPO terms than merely the set of terms chosen by our annotation pipeline on the basis of frequency and specificity of the term). Biocuration was performed by N.V., G.B., D.V., A.Z., M.H., and P.N.R., and all annotations were validated by P.N.R., who is both a computer scientist and a medical doctor. This allowed us to assess the true-positive, false-positive, and false-negative rates as shown in Tables S1–S41.