Integrating molecular dynamics and machine learning to identify potential apo-state conformational and solvent-exposure signatures associated with resistant KRAS mutants
TL;DR
A computational framework integrating molecular dynamics (MD)-derived structural, energetic, thermodynamic, and contact-based descriptors with machine learning may inform the design of inhibitors targeting secondary KRAS resistance mutations, pending validation in additional structurally independent mutant systems.
Abstract
Mutation-induced drug resistance is a major contributor to the failure of targeted cancer therapies, particularly in tumors driven by mutations in the KRAS oncogene. Although covalent inhibitors effectively target KRAS G12C, secondary mutations such as G12C/Y96C, G12C/Y96S, and G12C/Y96D confer resistance despite leaving the covalent attachment site intact. To investigate the conformational basis of this resistance, we developed a computational framework integrating molecular dynamics (MD)-derived structural, energetic, thermodynamic, and contact-based descriptors with machine learning. All simulations were performed in the apo (unbound) state; the results therefore reflect conformational and solvent-exposure correlates associated with resistant mutants rather than inhibitor-specific resistance mechanisms. Molecular descriptors extracted from MD simulations of treatment-sensitive and treatment-resistant KRAS systems were used to train logistic regression, random forest, support vector machine, and Bayesian network classifiers. To address the correlated nature of MD-derived conformers, model performance was evaluated using a system-independent mixed held-out validation scheme, and univariate analysis employed clustering-aware statistical approaches. Cross-referencing machine learning feature importance rankings with within-resistant-group variability testing revealed an important distinction between features reflecting mutant-identity-specific variation and those consistent with a shared resistance phenotype. Residue-level descriptors: solvent-accessible surface area variability at E62 and H95, Lennard-Jones 1,4 interaction energy, and root mean square fluctuation at M72 and H95 were both consistently discriminative across validation schemes and statistically consistent across all three resistant mutants, suggesting that they represent potential apo-state conformational and solvent-exposure signatures associated with resistant KRAS mutants. Our proof-of-concept workflow may inform the design of inhibitors targeting secondary KRAS resistance mutations, pending validation in additional structurally independent mutant systems.