Functional links between genes can be predicted using phylogenetic profiling, by correlating
the appearance and loss of homologs in subsets of species. However, effective genome-wide
phylogenetic profiling has been hindered by the large fraction of human genes related
to each other through historical duplication events. Here, we overcame this challenge
by automatically profiling over 30,000 groups of homologous human genes (orthogroups)
representing the entire protein-coding genome across 177 eukaryotic species (hOP profiles).
By generating a full pairwise orthogroup phylogenetic co-occurrence matrix, we derive
unbiased genome-wide predictions of functional modules (hOP modules). Our approach
predicts functions for hundreds of poorly characterized genes. The results suggest
evolutionary constraints that lead components of protein complexes and metabolic pathways
to co-evolve while genes in signaling and transcriptional networks do not. As a proof
of principle, we validated two subsets of candidates experimentally for their predicted
link to the actin-nucleating WASH complex and cilia/basal body function.