Fig. 4: Validation of variant effect prediction using literature-curated PTM-altering variants and proteogenomic datasets.
a, The sensitivity of DeepMVP in predicting PTM sites was assessed using the peptide with stronger experimental evidence for modification within a WT–variant (var) peptide pair, including WT peptide for variants known to decrease PTM (left) and var peptide for variants known to increase PTM (right). PTM sites were further categorized as overlapping with the variant position (direct, top) or located at a neighboring position relative to the variant (proximal, bottom). b, Directional concordance between effects predicted by DeepMVP and experimentally observed effects for the 191 variant–PTM pairs with correctly predicted PTM sites (from a). Pho, phosphorylation; Met, methylation; Sum, sumoylation; Ubi, ubiquitination (Ub); Gly, N-glycosylation; Ace, acetylation. c, Comparison of performance of DeepMVP and other tools on variant effect prediction, using two CPTAC cohorts (LSCC and UCEC). Predictions (increase or decrease) were separated into those supported (Supp., blue numbers) and conflicting (Conf., red numbers) on the basis of MS detection. This analysis was based on the first two batches of TMT experiments. d, Classification of DeepMVP’s predictions into the Supp. and Conf. groups on the basis of the complete CPTAC datasets.
