Realism and resolution: How AI-predicted protein structures differ from experimental structures
- 1KU Leuven
Abstract
Protein structure predictors can recover protein folds with remarkable accuracy, and their models are increasingly used as if they were experimental structures. Yet even for a confidently predicted fold, it remains difficult to judge how faithfully the model reproduces protein geometry. We introduce WhatRes, which assesses a structure from its coordinates on two complementary axes. Apparent resolution estimates the experimental resolution its local geometry most closely resembles. Realism measures how closely that geometry matches experimental structures at the same apparent resolution. Predicted models can score well on apparent resolution while showing systematically less variation in bond lengths and angles than is observed in high-resolution crystal structures. WhatRes reveals this over-idealisation even in high-confidence models.
Related
- Predicted and experimental protein-ligand coordinates with distance labels and evaluation splits
- Bioinformatic and experimental analyses revealed pathogen-derived pathoPEPs candidates predicted to control plant miRNAs expression and infection
- Systematic exploration of predicted quaternary structures within pandemic-relevant viral proteomes
- Predicted structure of the 89-kDa invasion-tip protein of Holospora obtusa, an endonuclear symbiont of Paramecium caudatum, suggests a pH-dependent conformational change
- Predicted condensation-related signatures localize to two C-terminal regions of the prion protein across cervid PRNP variants associated with chronic wasting disease
BibTeX
@misc{kant2026realism,
title = {{Realism and resolution: How AI-predicted protein structures differ from experimental structures}},
author = {R. van der Kant and J. Delgado Blanco and L. Serrano and J. Schymkowitz and F. Rousseau},
howpublished = {bioRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.06.764824},
url = {https://www.biorxiv.org/content/10.64898/2026.11.06.764824v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Realism and resolution: How AI-predicted protein structures differ from experimental structures
AU - R. van der Kant
AU - J. Delgado Blanco
AU - L. Serrano
AU - J. Schymkowitz
AU - F. Rousseau
PY - 2026
DA - 2026/11/06
PB - bioRxiv
DO - 10.64898/2026.11.06.764824
UR - https://www.biorxiv.org/content/10.64898/2026.11.06.764824v1
AB - Protein structure predictors can recover protein folds with remarkable accuracy, and their models are increasingly used as if they were experimental structures. Yet even for a confidently predicted fold, it remains difficult to judge how faithfully the model reproduces protein geometry. We introduce WhatRes, which assesses a structure from its coordinates on two complementary axes. Apparent resolution estimates the experimental resolution its local geometry most closely resembles. Realism measures how closely that geometry matches experimental structures at the same apparent resolution. Predicted models can score well on apparent resolution while showing systematically less variation in bond lengths and angles than is observed in high-resolution crystal structures. WhatRes reveals this over-idealisation even in high-confidence models.
ER -