Advancing Free Energy Methods for Drug Design

About This Grant

Project Summary/Abstract The Ding laboratory focuses on advancing drug design by developing computational methods that integrate molecular simulations with machine learning. We implement our innovative methods into open-source software, accompanied by comprehensive documentation and tutorials, making them accessible for both practical drug discovery programs and further methodological research. We believe that the full potential of computational techniques in drug design remains largely untapped, especially considering the rapid advancements in computing power and breakthroughs in machine learning and modeling technologies. Our long-term goal is to harness these advances to create transformative computational tools to accelerate drug discovery. In the next five years, we will focus on advancing methods for computing protein-ligand binding free energies, a key metric for drug design. Current techniques, such as relative binding free energy (RBFE) methods, have proven valuable in optimizing lead compounds. However, they are limited in scope, as they can only compute changes in binding free energy caused by small modifications on a reference compound. This restricts their use in exploring more chemically diverse compounds, which are often critical for early stages of drug discoveries. To address these limitations, we aim to advance methods for computing absolute binding free energies (ABFE), which allow for accurate binding affinity calculations without requiring a reference compound. While ABFE methods hold great promise, current approaches are either computationally expensive or inaccurate for wide practical use. Our goal is to create ABFE methods that overcome these challenges, making them both accurate and efficient for broad application in drug design. In parallel, we recognize that drug discovery is an iterative process in which computational predictions and experimental data accumulate over time, offering valuable insights into target protein and ligand interactions. However, current methods fail to fully leverage this expanding body of knowledge in a drug design program. To overcome this limitation, we will develop a Bayesian framework that incorporates prior information from both simulations and experiments into free energy calculations. This approach will enhance the precision, accuracy, and efficiency of these calculations, improving both relative and absolute free energy methods. During the development of these algorithms, we will rigorously evaluate their performance through benchmarking against existing datasets and collaborating with experimental groups in drug design projects. Additionally, we will test our method’s performance prospectively by participating in blind challenges. All new developments will be made available through open-source software, maximizing their impact on the broader drug discovery communities.