A Tool to Integrate User Expertise into Building Atomic Level Models for Large Biomolecular Systems

Abstract

The amount of available structural data for large macromolecular multi-protein complexes that carry out critical cellular processes have continuously increased in recent years. This increase is mainly due to advances in cryo-electron microscopy (cryo-EM). However, structure analysis still face serious challenges in dealing with highly flexible or multi-conformational domains; in fact structures of the latter are often inaccessible to experiments alone. To tackle these challenges computational modeling can be employed in combining structural data from cryo-EM, X-ray crystallography, and NMR spectroscopy. Instead of following the common strategy in macromolecular modeling, namely automating the process of computer aided structure analysis to avoid human bias, we integrate instead user expertise into an interactive version of model building. In our respective tool, Rosetta/MDFF, a user is guided by experimental data and automated structure prediction. Our approach enabled us to resolve the missing segments of mechanistically crucial subunits of the 26S proteasome, a 2.5 MDa multi-subunit molecular machine, which is a key player in protein degradation in cells. Incorporating user expertise into model building, in particular, is useful for a system, as complex as the proteasome since automated procedures may fail in dealing with the characterization of complicated, ambiguous structural regions. The interactive feature of Rosetta/MDFF allows a user to manipulate structures during molecular dynamics flexible fitting (MDFF) by manually pulling them to the desired regions of density.