Fold or flop: quality assessment of AlphaFold predictions on whole proteomes

Abstract Motivation Reliability of AlphaFold predictions is mainly assessed using the predicted Local Distance Difference Test (pLDDT). For model organisms, 30–40% of residues fall into the low-confidence pLDDT range. Moreover, pLDDT sometimes fails to flag physically implausible structures. This raises two questions: can more robust reliability indicators be identified, and do unreliable predictions share common structural or biophysical features? Results We use packing-based dimensionality reduction and clustering to assess the quality of whole-proteome predictions in the AlphaFold Database (AFDB), tracing connections with domain annotations, intrinsic disorder and stereochamical measures. We thus chart pathological structural motifs and unsupported disorder predictions, that reveal strengths and limitations of current self-assessment metrics and allow us to define a novel, misfold-aware structure quality assessment score. Availability The code to compute arity maps is available within the Structural Bioinformatics Library. See: AlphaFold analysis, and also Documentation, Applications, Installation guide. The code and data for rerunning analyses are made available at doi.org/10.5281/zenodo.18216693 Contact [email protected] online.