Entropy for End-Point and FFT-Based Binding Free Energy Calculations

项目总结

Project Summary Computers are often used to predict how tightly two molecules associate, their binding free energy. These predictions are helpful for designing drugs, predicting the consequences of genetic variation, and understanding how molecules interact to sustain life. Unfortunately, currently available methods are either fast or accurate, but not both. In general, fast methods do a poor job accounting for entropy, which is an important part of the free energy. The main objective of this project is to develop better ways to account for entropy in two popular techniques for studying molecular interactions: “end-point” simulations of the bound complexes and their unbound counterparts; and molecular docking based on the Fast Fourier Transform. Specifically, new ways to analyze calculation results will be derived, implemented, assessed, and optimized. Additionally, the methods will be combined with enhanced sampling techniques. Our new end-point and FFT-based methods will be assessed by their ability to reproduce benchmark results from slower but more accurate computational methods, as well as experimental results. The benchmark dataset will include protein-ligand complexes and protein-protein complexes with known binding affinities and crystal structures, as well as protein-protein complexes for which the effect of missense mutations on binding have been measured. We will also perform benchmark calculations on mutants of the tumor suppressor p53 that gain the ability to activate new proteins and promote tumor growth. In addition to serving as benchmarks, these calculations may provide mechanistic insight into how proteins bind various ligands and how p53 mutants gain new binding partners. Our new methods will also be tested in recurring community challenges: the “Drug Design Data Resource” (D3R) grand challenge to predict protein-ligand complex structures and affinities and the “Critical Assessment of PRediction of Interactions” (CAPRI) challenge for protein-protein structure prediction. These blinded challenges will allow for an unbiased comparison of our methods to those from other research groups. Finally, we will assess our methods in a drug discovery project. We will use established methods and our new methods to virtually screen a chemical library against a pair of structurally similar bacterial metabolic enzymes. One enzyme is relevant to active and the other to dormant bacteria. Compounds predicted to selectively bind the bacterial (opposed to human) enzymes will be experimentally tested in biochemical assays. We anticipate that our improved methods will be significantly more accurate than established approaches, advancing research ranging from interactome prediction to drug discovery.