Identifying Protein Conformational States in the PDB and Comparison to AlphaFold2 Predictions

Abstract

Proteins, as molecular machines, are necessarily dynamic macromolecules that carry out essential cellular functions. Recognising their stable conformations is important for understanding the molecular mechanisms of disease. While AI-based computational methods have enabled protein structure prediction, the prediction of protein dynamics remains a challenge. Here, we present a deterministic pipeline that clusters experimentally determined protein structures to comprehensively recognise conformational states across the Protein Data Bank. Our approach clusters protein chains based on a GLObal CONformation (GLOCON) difference score, which is computed from pairwise C-alpha distances. By superposing the clustered structures, differences and similarities in conformational states can be observed. Additionally, we offer users the ability to superpose predicted models from the AlphaFold Database to the clusters of PDB structures. This clustering pipeline significantly advances researchers ability to explore the conformational landscape within the PDB. All clustered and superposed models can be viewed in Mol* on the PDBe Knowledge Base website, or accessed in as raw annotations via our GraphAPI and FTP server. The clustering package is made available as an open-source Python3 package under the Apache-2.0 license.