Context-dependent protein structure prediction analysis and complex stoichiometry inference with MultimerMapper

Abstract

Advances in artificial intelligence (AI) have transformed the field of structural biology, notably with the accuracy level reached by AlphaFold in the prediction of protein structures. While cloud implementations have broadened access to these methods, there is still a lack of non-AI tools to systematically interpret and analyze the resulting data, especially under varying modeling contexts. Here, we present MultimerMapper, a pipeline and software suite designed to generate, extract, integrate, analyze, and visualize the behavior of multimeric systems in ensembles of overlapping protein structure predictions. It utilizes statistics, graph theory, and novel algorithms to interpret how protein-protein interfaces and protein conformations behave across different modeling contexts, under the premise that they vary depending on the components introduced into the predictions. Starting from a list of input sequences, a stoichiometric space exploration algorithm guides users in the generation of highly informative structure prediction ensembles. Through interactive visualizations, MultimerMapper highlights higher-order subassemblies, infers stoichiometries, and identifies alternative interaction modes and conformational changes statistically associated with the presence or absence of specific protein partners. MultimerMapper offers a new perspective on the dynamic nature of predicted protein complex structures, supporting researchers in the exploration of functional mechanisms and assembly paths.