FEP Protocol Builder: Optimization of Free Energy Perturbation Protocols Using Active Learning

Significant improvements have been made in the past decadeto methodsthat rapidly and accurately predict binding affinity through freeenergy perturbation (FEP) calculations. This has been driven by recentadvances in small-molecule force fields and sampling algorithms combinedwith the availability of low-cost parallel computing. Predictive accuraciesof & SIM;1 kcal mol(-1) have been regularly achieved,which are sufficient to drive potency optimization in modern drugdiscovery campaigns. Despite the robustness of these FEP approachesacross multiple target classes, there are invariably target systemsthat do not display expected performance with default FEP settings.Traditionally, these systems required labor-intensive manual protocoldevelopment to arrive at parameter settings that produce a predictiveFEP model. Due to the (a) relatively large parameter space to be explored,(b) significant compute requirements, and (c) limited understandingof how combinations of parameters can affect FEP performance, manualFEP protocol optimization can take weeks to months to complete, andoften does not involve rigorous train-test set splits, resulting inpotential overfitting. These manual FEP protocol development timelinesdo not coincide with tight drug discovery project timelines, essentiallypreventing the use of FEP calculations for these target systems. Here,we describe an automated workflow termed FEP Protocol Builder (FEP-PB)to rapidly generate accurate FEP protocols for systems that do notperform well with default settings. FEP-PB uses an active-learningworkflow to iteratively search the protocol parameter space to developaccurate FEP protocols. To validate this approach, we applied it topharmaceutically relevant systems where default FEP settings couldnot produce predictive models. We demonstrate that FEP-PB can rapidlygenerate accurate FEP protocols for the previously challenging MCL1system with limited human intervention. We also apply FEP-PB in areal-world drug discovery setting to generate an accurate FEP protocolfor the p97 system. FEP-PB is able to generate a more accurate protocolthan the expert user, rapidly validating p97 as amenable to free energycalculations. Additionally, through the active-learning workflow,we are able to gain insight into which parameters are most importantfor a given system. These results suggest that FEP-PB is a robusttool that can aid in rapidly developing accurate FEP protocols andincreasing the number of targets that are amenable to the technology.