Clinically Deployable Alchemical Free Energy Methods for Classifying Trimethoprim-Resistance Mutations

Abstract

Antimicrobial resistance (AMR) in Staphylococcus aureus and other clinically important pathogens has rendered first-line antifolate drugs such as trimethoprim (TMP) increasingly ineffective. Resistance is often driven by point mutations in the dihydrofolate reductase (DHFR) enzyme encoded by dfrB, dfrG, and dfrK. While whole-genome sequencing (WGS) has accelerated mutation discovery, the interpretation of novel or rare variants remains a major bottleneck, restricting the clinical utility of genomic diagnostics. Here, we investigate the application of alchemical free energy perturbation (FEP) and related atomistic molecular dynamics approaches for rapid, clinically deployable classification of TMP-resistance mutations. We present a computational pipeline that uses thermodynamic integration (TI), expanded ensemble methods, and absolute binding free energy calculations to estimate ΔΔG values associated with DHFR–TMP binding disruptions caused by single nucleotide polymorphisms (SNPs). Benchmarking against experimentally validated mutations reveals that alchemical free energy predictions can distinguish resistant from susceptible variants with high accuracy (AUC > 0.92). We discuss computational optimizations enabling sub-6-hour runtimes suitable for clinical microbiology laboratories, explore integration with machine-learning–guided priors, and evaluate the potential for real-time genomic AMR prediction. This work demonstrates that physics-based alchemical methods, once deemed too computationally expensive, are now viable as rapid, interpretable, and mechanistically faithful tools for classifying TMP resistance variants in clinical settings.