DRP-PSM: Multi-Level Feature Integration Reveals Hierarchical Contributions to Pathogenic Synonymous Mutation Prediction.

Synonymous mutations, a unique class of genetic variants, do not change the amino acid sequence of the encoded protein. Despite this, they can impact protein stability and function through diverse molecular mechanisms. Such subtle alterations can contribute to disease initiation and progression, making the prediction of pathogenic synonymous mutations crucial for understanding disease mechanisms as well as improving clinical diagnosis and treatment. In this study, we introduce DRP-PSM, a novel method for predicting pathogenic synonymous mutations that integrates DNA, RNA, and protein-level biological features. Building upon our earlier method, DRP-PSM greatly expands the feature set by incorporating protein-level information alongside enriched DNA and RNA signatures. Our goal is to elucidate how features from distinct biological levels contribute to the pathogenicity of synonymous mutations and to construct a comprehensive, multilevel prediction framework. DRP-PSM systematically integrates DNA-, RNA-, and protein-level sequence and structural features. Experimental results indicate that DNA-level features contribute the most to prediction accuracy, followed by RNA-level features, whereas protein-level features add only marginal utility. Notably, incorporating additional sequence- and structure-based descriptors yielded little performance gain, while biological features such as DNA conservation and splicing effect consistently dominated. These findings highlight that synonymous mutations primarily exert pathogenic effects through perturbations in splicing or transcriptional efficiency, rather than through translational or post-translational processes. This insight enhances our mechanistic understanding of their biological impact and underscores regulatory mechanisms as key targets for future therapeutic intervention.