Decoding Post-Translational Modification Patterns to Gain Functional Insights into Pathogenic Missense Variants

Genome sequencing has uncovered numerous pathogenic missense variants; however, their functional consequences remain largely unexplored, limiting our understanding of their precise roles in diseases. These variants may disrupt post-translational modifications (PTMs), which are crucial for cellular signaling and disease pathogenesis. Here, we present DeepVEP, a computational framework that uses deep learning-based PTM site prediction models to assess the impact of missense variants on six key PTMs. Our PTM site prediction models, trained on 397,524 PTM sites curated in PTMAtlas through systematic reanalysis of 241 PTM-enriched mass spectrometry datasets, significantly outperform existing models. DeepVEP’s variant effect predictions align closely with experimental results, as validated against literature-derived PTM-altering variants and two proteogenomic datasets. Its application to both pathogenic germline and somatic cancer variants creates a comprehensive landscape of PTM-altering disease variants. Furthermore, DeepVEP's interpretability facilitates connecting altered PTMs to potential modifying enzymes, opening new avenues for therapeutic interventions.