Computationally predicting binding affinity in protein–ligand complexes: free energy-based simulations and machine learning-based scoring functions

Abstract

Accurately predicting protein–ligand binding affinities can substantially facilitate the drug discovery process, but it remains as a difficult problem. To tackle the challenge, many computational methods have been proposed. Among these methods, free energy-based simulations and machine learning-based scoring functions can potentially provide accurate predictions. In this paper, we review these two classes of methods, following a number of thermodynamic cycles for the free energy-based simulations and a feature-representation taxonomy for the machine learning-based scoring functions. More recent deep learning-based predictions, where a hierarchy of feature representations are generally extracted, are also reviewed. Strengths and weaknesses of the two classes of methods, coupled with future directions for improvements, are comparatively discussed.

Computational studies of protein–drug binding affinity changes upon mutations in the drug target

“…For calculations of protein–ligand affinities ( K d , Δ G b ) from structural features, ML approaches might perform better than other statistics‐based scoring functions, but slower, physics based methods (MM/PBSA, MM/GBSA, LIE, FEP, TI, etc.) tend to be more accurate 107 . Yet, even at present, ML can be a useful tool to understand resistance mutations.…”

“…tend to be more accurate. 107 Yet, even at present, ML can be a useful tool to understand resistance mutations. To gain a realistic model of protein-ligand interactions in solution, MD simulations can be used to gain multiple data points from a single simulation, for example, protein-ligand interactions (number of contacts, different interaction-energies), surface area, orientation, and distances.…”

Computational Enzyme Engineering Pipelines for Optimized Production of Renewable Chemicals

“…Therefore, computational screening of enzyme stability, substrate-binding affinity, or activity can speed up the process of identifying the best candidates for experimental testing and simultaneously decrease the research expenses. Nevertheless, prediction of the exact biophysical properties of an enzyme in terms of changes in stability ( Pucci et al, 2018 ; Montanucci et al, 2019 ) and ligand-binding affinity ( Wang et al, 2021 ) upon mutation are still a great challenge and therefore require intensive computation.…”

“…Similarly, the precision of applications for predicting the impact of mutations on binding affinity is growing steadily. A study by Aldeghi et al (2018) compared different protein–ligand binding affinity prediction approaches based on a challenging benchmark set and showed that the Rosetta protocol ( flex_ddG ) ( Barlow et al, 2018 ), although originally developed for prediction of protein–protein binding affinity changes upon mutations, produced comparable results to computation-intensive free energy calculations ( Wang et al, 2021 ). A combination of both further increased the precision of the approach.…”

Protein pKa prediction by tree-based machine learning

“…In recent years, machine learning (ML) techniques have been applied to many scientific topics, including predictions of pKa values for non-protein molecules [88][89][90][91][92][93][94] and of protein-ligand binding affinity [95][96][97][98][99] . These topics are similar to pKa predictions because they both rely on predictions of free energy differences.…”