Presentation Title

Rapid end-state calculations for protein-ligand binding free energy

Start Date

12-11-2016 11:15 AM

End Date

12-11-2016 11:30 AM

Location

Watkins 2141

Type of Presentation

Oral Talk

Abstract

Computer-aided drug design (CADD) is an essential tool in the development of new pharmaceuticals but the computational cost of these methods makes them difficult to apply to emergent and neglected diseases. Protein-ligand binding affinity is a fundamental quantity predicted by CADD and the most accurate computational methods, such as thermodynamic integration, have demonstrated the potential for physics-based models to calculate this quantity. However, these methods often require enormous amounts of computing time, which can cause them to fail if large conformational changes in the protein occur. End-state methods, like molecular mechanics with generalized Born solvation (MM/GBSA), are another popular physics-based approach to calculate binding affinity. By simulating only the bound and unbound states, the method is much more computationally efficient but severe approximations reduce the accuracy. To develop a reliable, efficient end-state method, we first need to create a high quality benchmark dataset. For this we use the Community Structure-Activity Resource (CSAR) database. This consists of 115 high-quality protein-ligand crystal structures with binding affinity data. We have initially evaluated the dataset with a minimal end-state method, which uses a generalized Born solvent model and includes no conformational sampling. This provides a lower bound for accuracy and a null test with which to assess the different approximations and protocols. While the absolute binding free energies calculated via this minimal method had large errors, the Pearson correlation coefficient was R=0.60. This indicates that while there is much room for improvement, even this simple method has some success predicting relative binding affinities.

This document is currently not available here.

Share

COinS
 
Nov 12th, 11:15 AM Nov 12th, 11:30 AM

Rapid end-state calculations for protein-ligand binding free energy

Watkins 2141

Computer-aided drug design (CADD) is an essential tool in the development of new pharmaceuticals but the computational cost of these methods makes them difficult to apply to emergent and neglected diseases. Protein-ligand binding affinity is a fundamental quantity predicted by CADD and the most accurate computational methods, such as thermodynamic integration, have demonstrated the potential for physics-based models to calculate this quantity. However, these methods often require enormous amounts of computing time, which can cause them to fail if large conformational changes in the protein occur. End-state methods, like molecular mechanics with generalized Born solvation (MM/GBSA), are another popular physics-based approach to calculate binding affinity. By simulating only the bound and unbound states, the method is much more computationally efficient but severe approximations reduce the accuracy. To develop a reliable, efficient end-state method, we first need to create a high quality benchmark dataset. For this we use the Community Structure-Activity Resource (CSAR) database. This consists of 115 high-quality protein-ligand crystal structures with binding affinity data. We have initially evaluated the dataset with a minimal end-state method, which uses a generalized Born solvent model and includes no conformational sampling. This provides a lower bound for accuracy and a null test with which to assess the different approximations and protocols. While the absolute binding free energies calculated via this minimal method had large errors, the Pearson correlation coefficient was R=0.60. This indicates that while there is much room for improvement, even this simple method has some success predicting relative binding affinities.