On Free Energy Calculations in Drug Discovery.

This Account discusses recent progress and challenges in binding free energy computations, focusing on two classes of enhanced sampling techniques: alchemical transformations and path-based methods. Binding free energy is a crucial metric in drug discovery, as it measures the affinity of a ligand for its target receptor. Free energy and affinity guide the ranking and selection of potential drug candidates. The theoretical foundations of free energy calculations were established several decades ago, but their efficient application to drug-target binding remains a grand challenge in computational drug design. The main obstacles stem from sampling issues (as binding is a rare event), force field accuracy limitations, and simulation convergence. Alchemical transformations are now the most used methods for computing binding free energies in the pharmaceutical industry. However, while they efficiently calculate energy differences, the application of these methods is often limited to relative binding free energy calculations. Absolute and accurate (error < 1 kcal/mol) binding free energy predictions remain one of the great challenges for computational chemists and physicists. Another limitation of alchemical methods is that they lack the ability to provide mechanistic or kinetic insights into the binding process, crucial for optimizing lead compounds and designing novel therapies. Path-based methods offer, in principle, the possibility to accurately estimate absolute binding free energy while also providing insights into binding pathways and interactions.This Account explores recent advances in binding free energy methods for drug-target recognition and binding. In particular, we discuss the similarities and differences between alchemical and path-based approaches, highlighting recent innovations in both families of methods and providing perspectives from our group's contributions. We examine the foundational role of alchemical methods, which have been employed since the inception of free energy calculations, in both equilibrium and nonequilibrium contexts. We also emphasize the growing importance of path-based methods in drug discovery and their ability to predict binding and unbinding pathways, free energy profiles, and binding free energy estimates. In particular, the combination of path methods with machine learning has proven to be a powerful means for accurate path generation and free energy estimations. Building on our recent research, we discuss several path-based applications for drug discovery. Moreover, we focus on two semiautomatic protocols representing our group's state-of-the-art in free energy calculations. The first protocol is based on MetaDynamics simulation. From this, a recent innovation is instead based on nonequilibrium simulations combined with nonequilibrium estimators. We discuss in depth the advantages and drawbacks of equilibrium and nonequilibrium approaches to drug-target binding free energy predictions.