Assessing Structural Classification Using AlphaFold2 Models Through ECOD-Based Comparative Analysis.

ABSTRACT: Identifying homologous proteins is a fundamental task in structural bioinformatics. While AlphaFold2 has revolutionized protein structure prediction, the extent to which structure comparison of its models can reliably detect homologs remains unclear. In this study, we evaluate the feasibility of homology detection using AlphaFold2-predicted structures through structural comparisons. We considered the classification of the ECOD database for experimental structures as the correct standard and obtained their corresponding predicted models from AlphaFoldDB. To ensure blind assessment, we divided the structures into test and train sets according to their release date. Predicted and experimental 3D structures in the test and train sets were compared using 3D structure comparisons (MATRAS, Dali,

SUBMITTER: Kawabata T 

PROVIDER: S-EPMC12314581 | biostudies-literature | 2025 Sep

REPOSITORIES: biostudies-literature