Improving membrane protein modeling and design using empirical data

Abstract

Membrane proteins are an important class of proteins that represent approximately 30% of the open reading frame. However, due to their inherent flexibility, membrane proteins are difficult to structurally characterize, resulting in limited available data. Computational methods are able to leverage existing information to create accurate predictions. Over the past few decades, additional experimental studies evaluating the thermostability of single point mutations has become more readily available. After the initial benchmark of the computational design of membrane proteins, it was found that there was a bias for designing in Leucine. Experimentally derived mutation-induced stability changes were utilized to further evaluate the membrane protein energy function. Regression analysis revealed that the membrane score terms were indeed important for accurate predictions that approximate experimentally derived changes in energy between the mutant and wild-type states. The results of these studies have many applications that include design of a stable membrane protein scaffold, design for stabilizing membrane proteins for structural characterization, modeling of membrane proteins variants, and improving characterization of variants of unknown significance.