Benchmarking Collective Motion Predictions of Elastic Network Models

Abstract

Elastic Network Models (ENMs) provide approximate descriptions of the collective conformational freedom intrinsically accessible to protein structures. Despite their simplicity, many ENMs have been repeatedly demonstrated to be in qualitative agreement with experimental observations of protein conformational changes. A wide variety of ENMs has been proposed in the last two decades, and they have become a favored approach for elucidating the collective components of protein motion. Since experimental data on the temporal correlations of atoms in proteins are currently unavailable, most attempts at quantitative assessment of ENMs rely on comparisons of predicted atomic fluctuations with crystallographic B-factors. Using such benchmarks, diverse ENMs have been shown to give very similar predictions. In contrast to this, we have revealed consistent differences between ENMs by comparing their inter-atomic covariances with those obtained from Molecular Dynamics sampling.Some of us recently demonstrated the importance of considering the covariance structure explicitly when comparing the conformational freedom of protein homologs (Fuglebakk et.al., Bioinformatics 2012) . In the same work we developed rigorous means of carrying out such comparisons. I will present results obtained from applying this approach to do a thorough assessment of a selection of ENMs. We find consistent differences between the ENMs. However, the differences do not always follow a priori expectations based on the models complexity. Rather, even some of the simpler ENMs give good approximations to the covariances obtained from Molecular Dynamics. Often this comes at the cost of less reliably predicting the atomic fluctuations that have commonly been used for parameterization and validation. Lastly, we find that the use of B-factors for parameterization or validation warrants particular care, as the common approach for doing this favors ENMs that constrain collective motion.