We then examined gene families that were not predictive of PCP identity. ~243 families with AUROC<0.6 (Table S5) were mostly implicated in generic cell functions ranging from DNA replication to energy conversion etc. This analysis also revealed gene families that a priori might be assumed to predict PCP identity but did not. These include 1) a substantial set of cytoplasmic signaling kinases (e.g. the entire mitogen-activated kinase cascade), phosolipases (30 members); 2) most cytoskeleton components (actins, myosin, dyneins, kinesins); 3) core components of vesicle release machinery (Table S5). These non-predictive gene families contrasted the top 40 predictive families in two major properties: they were all cytosolic instead of membrane or membrane proximal; and they often made up the generic scaffold of macromolecular complexes or pathways instead of the key regulatory components.
