Welcome to the Computational Chemical Biology Lab

Spectroscopic methods have shown that ligands perturb the equilibria of 7TMR intracellular pocket conformations, but the identity of these conformations has remained unclear. We have developed a method that combines molecular simulation and machine learning to identify conformations and their signaling efficacies.

The standard protocol for analysis of concentration response curves involves normalization by the control data but not fitting to it. We showed that including control data in the fit improves the accuracy and precision of parameter estimates.

We trained deep neural networks to map between peptide conformations. The learned mappings were used to estimate free energy differences.

Although it is well-known that ligands are polarized by proteins, the magnitude of this effect had not been quantified in many systems. We evaluated the ligand polarization energy for several hundred protein-ligand complexes and showed that it is a large and highly variable component of the binding energy.

The binding potential of mean force (BPMF), the binding free energy between a flexible ligand and a rigid receptor, is a critical ingredient for esimating binding free energies with implicit ligand theory. We have written software to precisely estimate this quantity and tested it on a diverse set of 85 protein-ligand complexes.

Although there are many ways to select representative snapshots from a molecular dynamics simulation to perform molecular docking, it has been unclear how to assess these methods. We pointed out that this procedure is an example of a statistical method, stratified sampling, and that the efficiency of stratification can be used to assess ensemble reduction methods.

We have shown that the Fast Fourier transform can be used to calculate binding free energies using implicit ligand theory. Previously, the use of FFT in molecular docking was only to estimate interaction energies.

We have identified the ubiquinone binding site in the bacterial ion pump NQR. This binding site is not obvious from the crystal structure. It is a possible target for structure-based drug design.

See article in IIT Today.

We have shown how to use constrained molecular dynamics, such as torsional dynamics, as a Monte Carlo move for molecular simulation. Previously, molecular simulations based on constrained dynamics would not sample from the appropriate distribution or not sample the entirely of configuration space.

See article in IIT Today.

We have performed large-scale protein-ligand binding free energy calculations using implicit ligand theory. Previously, implicit ligand theory was only applied to host-guest systems.

We have shown that if there are a sufficient number of states in a replica exchange simulation, the precise definition of states does not affect sampling efficiency. Previously, many scientists thought that the number of states in replica exchange should be carefully optimized.