To identify putatively functional variation, we applied four popular methods applicable to non-synonymous variants (PolyPhen2, SIFT, a likelihood ratio test, and MutationTaster) and three conservation-based methods applicable to all types of variants [GERP, PhyloP, and a novel population genetics approach that combines conservation information with the SFS that we designate SFS-Del (18)]. About 47% of all SNVs (74% of nonsynonymous and 6% of synonymous variants) are predicted to be deleterious by one or more method (Fig. 1A); however, overlap among methods is modest. For example, only 1% of nonsynonymous variants are predicted to be functional by all seven methods, and variants predicted by any single approach are likely to have a high false-positive rate (Fig. 1A). Therefore, we used a conservative majority rule approach and deemed nonsynonymous variants predicted by at least four of the seven applicable methods and synonymous sites predicted by at least two of the three applicable methods (fig. S13) to be putatively functional. In total, 16.9% of SNVs (85,224) meet this criteria, of which 81,170 were nonsynonymous SNVs. About 95.7% (81,555) of all SNVs conservatively predicted
