Prediction of Recurrence-Defined Candidate Driver Synonymous Variants Through Integrative Modeling of Genetic Regulatory Features

Synonymous mutations influence RNA splicing, mRNA stability, and translational efficiency without altering amino acid sequences, yet accurately identifying those that drive cancer remains a persistent challenge in cancer genetics. Current computational approaches are constrained by limited feature representation, a scarcity of experimentally validated data, and the diverse regulatory processes potentially associated with the functional relevance of synonymous variants. This study aimed to overcome these limitations by developing an integrative predictive framework. To address the limitations of existing methods, we developed an Integrative Genetic Regulatory Model (IGRM), an integrative framework based on genetic regulatory features, for prioritizing recurrence-defined candidate driver synonymous mutations. Guided by key regulatory processes in genetic information flow, IGRM incorporates six categories of functional features related to transcriptional regulation, RNA processing, and translation, together with RNA-binding protein binding sites, RNA modification sites, and multiple pathogenicity scores. Moreover, IGRM adopts a stacked ensemble learning strategy to enhance predictive performance and generalizability. Across two held-out test sets, IGRM achieved higher AUC and AUPR values than the evaluated methods in distinguishing recurrence-defined positive and negative synonymous variants. Local interpretation of top-ranked candidate variants revealed that computationally predicted RBP-related features were key contributors to their prioritization, generating testable hypotheses about their regulatory relevance. IGRM provides an integrative computational approach for prioritizing recurrence-defined candidate driver synonymous variants and examining the annotations contributing to model predictions.