Prediction of protein assemblies by structure sampling followed by interface-focused scoring

Abstract

Proteins often function as part of permanent or transient multimeric complexes, and understanding function of these assemblies requires knowledge of their three-dimensional structures. While the ability of AlphaFold to predict structures of individual proteins with unprecedented accuracy has revolutionized structural biology, modeling structures of protein assemblies remains challenging. To address this challenge, we developed a protocol for predicting structures of protein complexes involving model sampling followed by scoring focused on the subunit-subunit interaction interface. In this protocol, we diversified AlphaFold models by varying construction and pairing of multiple sequence alignments as well as increasing the number of recycles. In cases when AlphaFold failed to assemble a full protein complex or produced unreliable results, additional diverse models were constructed by docking of monomers or subcomplexes. All the models were then scored using a newly developed method, VoroIF-jury, which relies only on structural information. Notably, VoroIF-jury is independent of AlphaFold self-assessment scores and therefore can be used to rank models originating from different structure prediction methods. We tested our protocol in CASP15 and obtained top results, significantly outperforming the standard AlphaFold-Multimer pipeline. Analysis of our results showed that the accuracy of our assembly models was capped mainly by structure sampling rather than model scoring. This observation suggests that better sampling, especially for the antibody-antigen complexes, may lead to further improvement. Our protocol is expected to be useful for modeling and/or scoring protein assemblies.

The impact of AI‐based modeling on the accuracy of protein assembly prediction: Insights from CASP15

“…[76][77][78] After pooling all models, they ranked the models with their consensus VoroIF-jury method. 79 Wallner sampled an exhaustive set of models by using AF2M versions 2.1 25 and 2.2. 32 While using these AI models, they activated the dropout layers in the Evoformer block, increased the number of recycles, as well as the number of seeds, and turned on/off the template usage.…”

A Perspective on the Prospective Use of AI in Protein Structure Prediction

“…Other ideas correspond to tools that do not exist yet but could be developed in the near future, in particular given the release of the TRILL toolkit for interfacing existing tools, which in turn could facilitate the contribution of experimentalists: A meta-tool that combines all modeling programs in one pass, ranks them globally and suggests an ideal composite model. Although a jury method has been proposed for the prediction of protein–protein complexes, this is still a largely unexplored possibility. A predictor of post-translational modifications. All-atom versions of RoseTTAFold and AlphaFold2 have been announced very recently and should unlock this current limitation. A predictor of function. …”

VoroIF-GNN: Voronoi tessellation-derived protein-protein interface assessment using a graph neural network

“…VoroIF-GNN was also used by "Venclovas" group in two ways: for the second global score, and as a major component of the newly developed VoroIF-jury scoring method. VoroIF-jury is described in the accompanying publication dedicated to the methodology and results of the protein assembly modeling by the "Venclovas" group [11].…”

“…For reference, under the name of "VoroMQA-select-2020" we also included our CASP14 EMA method. Since VoroIF-jury is described in an accompanying paper in this Proteins issue [11], here we focus exclusively on VoroIF-GNN. We describe the underlying ideas and their implementation as well as provide a concise description of its performance in the EMA category of CASP15.…”