AI told you so: navigating protein structure prediction in the era of machine learning

Abstract

Every protein in our cells has evolved to fold into a specific structure to perform its functions. However, determining these structures experimentally is often challenging, in some cases taking years. Recently, machine-learning algorithms have been designed to predict a protein’s structure directly from its amino acid sequence in minutes to hours. Since the release of the first of these algorithms, AlphaFold and RoseTTAFold, several more have been developed. These have been complemented by tools that leverage the outputs to give structural context to biochemical data, screen for novel protein–protein interactions or even help solve experimental structures. In addition, several public resources have incorporated the predictions into their databases, making the data open to all. Here, we provide a user-focused perspective on machine-learning protein structure prediction, covering some of the popular applications and highlighting caveats. Used effectively, predictive programs offer the potential to speed up research and guide experimental design.