Protein Detective

Due to advances in Cryo-Electron Microscopy (Cryo-EM) hardware and software, we can characterize biomolecules and their interacting partners in their native environment through Cryo-Electron Tomography (Cryo-ET). To visualize proteins at an intermediate resolution range (5-10 Å) model-fitting software is commonly used. One major limitation of these methods is, however, that the identity of the proteins to be modelled needs to be known. This project aims at developing a software pipeline to identify unassigned densities in intermediate-resolution maps or poorly-resolved densities and subsequently build reliable structural models. The unknown proteins will be identified by selecting a pool of proteins from the AlphaFold2 database, filtered based on cellular location, size, and secondary structure elements. We will then use a fast rigid-body fitting tool (e.g. our in-house PowerFit, which will be further optimised for efficiency) to fit each selected protein into the unknown density to determine the most likely candidate. The identification process will be further refined by modelling the full complex, considering the interaction energetics between components using HADDOCK3, and AI-based scoring methods such as our DeepRank software. The resulting protein identification and model-building pipeline harvesting AI-generated structures will provide the community with a new tool to build realistic models of complexes for yet unidentified protein components in medium-resolution Cryo-ET maps. Considering the size of the search problem, efficient software needs to be developed to search and filter public databases, and then fit the selected models against the EM data to identify possible candidates.