The maximal and current accuracy of rigorous protein-ligand binding free energy calculations

It is now well recognized that computational techniques can greatly speed up the identification of hits and accelerate the optimization of such hits to lead series and development candidate molecules. In particular, a class of rigorous physics-based methods known as free energy perturbation (FEP) have emerged as the most consistently accurate relative affinity prediction tool available to support such efforts. Yet, there still remains uncertainty about how accurate these techniques are, and indeed, how accurate they can ever be. In this study, we assemble what we believe to be the largest publicly available data set of proteins and congeneric series of small molecules to date and assess the accuracy of the leading FEP workflow in predicting relative binding affinities. To ascertain the limit of achievable accuracy, we survey the reproducibility of experimental relative affinity measurements by comparing chemical series that have been assayed by two or more different techniques. We find a wide variability in experimental accuracy and a general correspondence between binding and functional assays. When the protein and ligand structures from our data set are prepared with care, we find that FEP can achieve a level of accuracy close to what we find in our experimental survey. Throughout, we highlight reliable protocols that can help maximize the accuracy of FEP in prospective studies.