How good are ensembles in improving QSAR models? The case with eCoRIA

A conceptually new idea in quantitative structure–activity relationships (QSAR) which makes use of ensembles from molecular dynamics (MD) trajectories and information retrieved from enzyme–inhibitor binding thermodynamics is presented in this study. This new methodology, termed ensemble comparative residue interaction analysis (eCoRIA), attempts to overcome the current one chemical–one structure–one parameter value dogma in computational chemistry by modeling the biological activity as a function of molecular descriptors derived from an ensemble of conformers of enzyme–inhibitor complexes. The approach is distinctly different from the standard QSAR methodology which uses a single low-energy conformation or the properties averaged over a set of conformers to correlate with the activity. Each conformational ensemble derived from MD simulations is analyzed for the distribution of the non-bonded interaction energies (steric, electrostatic, and hydrophobic) along with solvation, strain, and entropic energy of the inhibitors with the individual amino acid residues in the receptor and these are correlated to the activity through a QSAR model. The scope of the new method is demonstrated with three diverse enzyme–inhibitor data-sets – glycogen phosphorylase b, human immunodeficiency virus-1 protease and cyclin-dependent kinase 2. The QSAR equations derived from the methodology have revealed all the structure activity relationships previously reported for these classes of molecules as well as uncovered some features that were hitherto unknown and may have a hidden role in driving the ligand–receptor-binding process. Impressive improvements in the predictions of affinity have been achieved compared to other QSAR formalisms namely CoMFA, CoMSIA (receptor-independent QSAR techniques), and CoRIA (a receptor-dependent QSAR technique). eCoRIA could provide an understanding of the thermodynamic properties influencing the ligand–receptor binding over a time scale as sampled by the MD simulation. The advantage of analyzing enzyme–inhibitor interaction energies in a statistical domain is that the noise due to inaccuracies in the potential energy functions can be reduced and mechanistically important interaction terms related to protein–ligand binding specificity can be identified which can assist the medicinal chemists in designing new molecules and biologists in studying the influence of position-specific mutations in the receptor on ligand binding.