On the relationship between docking scores and protein conformational changes in HIV-1 protease.

We present a detailed investigation of the effect of the crystallographic structure of the HIV-1 protease (PR) on the binding energy of different classes of inhibitors obtained from docking simulations. The crystal structures of 222 HIV-1 proteases (in wild-type and mutant forms) and 202 inhibitors were downloaded from appropriate databases. A cross-docking approach (docking of all 202 inhibitors to all 222 PR structures) using Autodock Vina was implemented. The protease structures were clustered using a Kohonen self-organization map analysis of the data matrix of docking energies. The obtained clusters of PRs were correlated with the x-y-z coordinates of the PR structures to identify structural segments underlying this clustering. The PR structures were clustered into 4 classes. One of these classes exhibits rather strong binding with almost all inhibitors, while another class exhibits rather weak binding. The remaining two classes are intermediate in binding strength. The selectivity ratio indices for the carbon-alpha atoms of the PR structures indicate that conformational motion of residues outside the binding pocket contributes significantly to the discrimination of the 4 classes.