Variant effect predictions capture some aspects of deep mutational scanning experiments

Abstract

Deep mutational scanning (DMS) studies exploit the mutational landscape of sequence variation by systematically and comprehensively assaying the effect of single amino acid variants (SAVs) for particular proteins. Different experimental protocols proxy effect through a diversity of measures. We evaluated three early prediction methods trained on traditional variant effect data (PolyPhen-2, SIFT, SNAP2) along with a regression method optimized on DMS data (Envision). On a common subset of 32,981 SAVs, all methods capture some aspects of variant effects, albeit not the same. Early effect prediction methods correlated slightly more with measurements and better classified binary states (effect or neutral), while Envision predicted better the precise degree of effect. Most surprising was that a simple approach predicting residues conserved in families (found and aligned by PSI-BLAST) in many cases outperformed other methods. All methods predicted beneficial effects (gain-of-function) significantly worse than deleterious (loss-of-function). For the few proteins with several DMS measurements, experiments agreed more with each other than predictions with experiments. Our findings highlight challenges and opportunities of DMS for improving variant effect predictions.