Active Learning Guided Drug Design Lead Optimization Based on Relative Binding Free Energy Modeling

Abstract

In silico identification of potent protein inhibitors commonly requires prediction of a ligand binding free energy (BFE). Thermodynamics integration (TI) based on molecular dynamics (MD) simulations is a BFE calculation method capable of acquiring accurate BFE, but it is computationally expensive and time-consuming. In this work, we have developed an efficient automated workflow for identifying compounds with the lowest BFE among thousands of congeneric ligands, which requires only hundreds of TI calculations. Automated machine learning (AutoML) orchestrated by active learning (AL) in an AL–AutoML workflow allows unbiased and efficient search for a small set of best-performing molecules. We have applied this workflow to select inhibitors of the SARS-CoV-2 papain-like protease and were able to find 133 compounds with improved binding affinity, including 16 compounds with better than 100-fold binding affinity improvement. We obtained a hit rate that outperforms that expected of traditional expert medicinal chemist-guided campaigns. Thus, we demonstrate that the combination of AL and AutoML with free energy simulations provides at least 20× speedup relative to the naïve brute force approaches.

Recent Advances in Alchemical Binding Free Energy Calculations for Drug Discovery

“…Iteratively training automated ML (AutoML) models with a limited number of RBFE calculations has been suggested as an alternative. Gusev et al 33 proposed an iterative AutoML workflow where Amber GPU-TI RBFE calculations were conducted for a small number of SARS-CoV-2 papain-like protease binders. For a focused set of 8175 potential ligands, the centroids of 45 clusters were subjected to Amber GPU-TI RBFE calculations.…”

Keeping pace with the explosive growth of chemical libraries with structure‐based virtual screening

“…By screening a fraction of a large chemical library and leveraging AL to predict the remainder, authors were able to identify phosphodiesterase 2 inhibitors with high predicted binding affinity. Similarly, Gusev et al proposed an AL strategy to accelerate RBFE calculations based on thermodynamic integration (TI) simulations 103 . For each AL cycle, their AutoML platform selects the best performing model (among trained RF, MLP, linear regression, SVM, and GPR) as a surrogate for computationally expensive TI calculations.…”

“…Similarly, Gusev et al proposed an AL strategy to accelerate RBFE calculations based on thermodynamic integration (TI) simulations. 103 For each AL cycle, their AutoML platform selects the best performing model (among trained RF, MLP, linear regression, SVM, and GPR) as a surrogate for computationally expensive TI calculations. Their method predicted potential high-affinity SARS-CoV-2 PLpro inhibitors by running TI simulations for only 3% of a focused library containing 8000 compounds.…”

Benchmarking Active Learning Protocols for Ligand-Binding Affinity Prediction

“…In these cases, 100s to 1000 compounds are selected out of pools containing up to 100,000 samples. The sheer size of the compound pool goes hand in hand with a high degree of diversity compared to low-throughput use cases, putting more strain on the AL pipeline and necessitating a careful selection of molecular features, ML models, and acquisition methods. ,,, In addition to the challenge posed by data set sizes and diversity, using RBFEs or docking scores in lieu of experimental binding affinities introduces errors of systematic and stochastic nature, which are often not well characterized in advance.…”