Speaker
Description
Abstract:
Missense mutations in oncogenic proteins such as PTEN, PI3Kα, MEK1/2, SHP2, and RAS can lead to remarkably different phenotypes, including cancer and neurodevelopmental disorders (NDDs). A key challenge is to explain why mutations in the same protein result in divergent clinical outcomes. Here, we present integrative machine learning (ML) and structural dynamics approaches to address this problem. The first part introduces protPheMut, an interpretable ML tool that integrates sequence, biophysical, and network-dynamics features, enabling accurate classification of cancer- versus NDD-associated mutations across multiple proteins. Compared with existing predictors, protPheMut achieves superior accuracy and provides mechanistic insights through SHAP-based model explanations. The second part focuses on PTEN, a paradigmatic gene shared by cancer and autism spectrum disorder (ASD). By combining ML with molecular dynamics simulations, we reveal that cancer-associated PTEN mutations often cause global destabilization and long-range allosteric perturbations, whereas ASD- and dual-phenotype mutations induce localized dynamic changes in conserved functional regions. Together, these findings highlight the power of combining ML and protein dynamics to bridge genotype–phenotype relationships and to uncover shared molecular mechanisms, thereby informing precision medicine strategies for both cancer and NDDs.
References:
[1] J Chem Inf Model. 11; 65(15): 8375-8384.
[2] J Chem Inf Model. 28; 65(8): 4173-4188.
