Automated Mining of Hinge-like Protein Modules from AlphaFold PAE: BCR-Parts

Keywords: AlphaFold; Predicted Aligned Error (PAE); block contrast ratio (BCR); hinge detection; permutation testing; BH-FDR; molecular robotics; protein parts library

TL;DR: We mine hinge-like protein parts directly from AlphaFold PAE. A simple BCR statistic plus two-sided calibration yields a heavy right tail and 314 BH discoveries out of 1,476 rows. External RMSD evidence is left for future re-runs.

Abstract: We introduce BCRParts, a simple pipeline that turns single-structure predictions into candidates for mechanically separable protein “parts”. The method segments the AlphaFold PAE into k in {2,3} blocks, scores block–contrast (BCR), and assigns significance by permutation p-values with Benjamini–Hochberg FDR applied once per run. On a curated cohort we processed 1,476 targets in parallel. The score distribution shows a clear right tail, and—importantly—the BCR-derived score correlates with an independent PAE diagnostic, the asymmetry index (Pearson r=0.59, Spearman rho=0.69. Using a two-sided calibration (see Methods), 314/1,476 rows pass BH at $q\le0.05$ ; top candidates and attributes (e.g., hinge length, opening angle) are summarized in the Table. External evidence (PDBFlex/CoDNaS) could not be robustly retrieved in time for this submission; we therefore report coverage diagnostics and leave full cross-dataset validation to future work.

Submission Number: 306