Conformational Sampling Reveals Amino Acids with a Steric Influence on Specificity

Abstract:

Flexible representations of protein structures can enable structure comparison algorithms to findremotely homologous proteins, even when they have been crystallized in different conformations. By compensating for large spatial variations, these representations can enable these algorithms to better detect remote similarities in the space of protein structures. Subtle variations in protein structures can also have a substantial impact structure comparison. For example, the motion of a single side chain into a binding cavity can make the cavity appear totally dissimilar to identical binding sites, even though, in reality, the presence of the side chain does not affect binding. To address the impact of subtle conformational variations, this paper describes FAVA (Flexible Aggregate Volumetric Analysis), an algorithm that enables comparisons of ligand binding sites while compensating for subtle, localized flexibility. FAVA integrates hundreds of conformational samples, sourced from any molecular simulation software that provides all-atom detail, to characterize the geometry of ligand binding sites as they frequently appear. This representation enables rare conformations, as defined by the user, to be excluded from the structural comparison. In our results, on three families of serine proteases and three families of enolases, we show that despite substantial binding site variations, FAVA is able to correctly classify families with different binding preferences. We also demonstrate that FAVA can examine the motion of individual amino acids to identify those that influence ligand binding specificity. Together, these capabilities demonstrate that comparison errors associated with small conformational variations, which can substantially alter the geometry of ligand binding sites and other local features, can be mitigated by an analysis of many conformational samples.