FEP Ω: The End of Parameter Tuning

Free energy perturbation (FEP) is the gold standard in structure-based drug design, but its lack of out-of-the-box accuracy requires extensive parameter tuning, often delaying its use in compound design. To address this, we present FEP Ω, a novel machine learning native platform for FEP that directly resolves these limitations and accelerates deployment. FEP Ω flips the traditional paradigm of free energy perturbation on its head by eliminating a priori force field adjustments, alchemical intermediates, and network corrections. Instead, FEP Ω uses a standardized and automated setup protocol while leveraging machine learning post-simulation to achieve unprecedented, data-driven accuracy in a streamlined workflow. Together, these innovations make FEP lighter, faster, more accurate, and immediately practical for hit-to-lead and lead optimization. Finally, benchmarking against Schrödinger’s FEP-PB demonstrates superior accuracy in a fraction of the time, removing longstanding barriers to FEP adoption in drug discovery. Prospective virtual screening with FEP Ω identified novel dipeptidyl peptidase-4 chemical matter with experimental potencies exceeding that of sitagliptin (Januvia), achieving a 100% hit rate across all ten compounds tested.