Leveraging state-of-the-art AI structure predictors to generate atomic resolution ensembles of intrinsically disordered proteins
- 1Institut Pasteur
Abstract
Intrinsically disordered proteins (IDPs) are central to many biological processes, but their function depends on dynamic conformational ensembles rather than a single structure. Despite the impact of artificial intelligence on protein structure prediction, determining these ensembles at atomistic resolution remains challenging. Recently, we introduced bAIes, a framework that leverages the structural information encoded in AlphaFold2 distograms to generate conformational ensembles of IDPs at atomistic resolution. Here, we extend this framework to six state-of-the-art structure prediction tools (AlphaFold2, AlphaFold3, Boltz-1, Boltz-2, ESMFold, and ESMFold2) and systematically compare the structural information encoded in their predicted inter-residue distance distributions. Despite differences between predictors, we show that bAIes generates ensembles with comparable accuracy against experimental NMR and SAXS data, and competitive performance relative to state-of-the-art all-atom simulations and alternative IDP ensemble generation methods. These results demonstrate that bAIes is robust to the choice of structure predictor and that the distance distributions generated by different deep learning architectures contain valuable information about the conformational preferences of disordered proteins. More broadly, our work highlights the potential of combining AI-derived structural information with physics-based modeling to generate atomistic ensembles of intrinsically disordered proteins and regions.
Related
- Single residues within the YTHDF intrinsically disordered region control m6A-dependent mRNA decay
- An intrinsically disordered proline-rich domain acts in synergy with AGP-type glycosylation to control the protein fate of FLA4
- Functional importance and paralog-specificity of the intrinsically disordered RsmE C-terminus region in Pseudomonas fluorescens Pf0-1
- Intrinsically disordered regions navigate DNA replication licensing factors to specific genomic loci.
- Intrinsically disordered regions influence protein motif usage in Drosophila Ubx and AbdA Hox proteins
BibTeX
@misc{schnapka2026leveraging,
title = {{Leveraging state-of-the-art AI structure predictors to generate atomic resolution ensembles of intrinsically disordered proteins}},
author = {V. Schnapka and N. Portal and M. Bonomi},
howpublished = {bioRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.10.765624},
url = {https://www.biorxiv.org/content/10.64898/2026.11.10.765624v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Leveraging state-of-the-art AI structure predictors to generate atomic resolution ensembles of intrinsically disordered proteins
AU - V. Schnapka
AU - N. Portal
AU - M. Bonomi
PY - 2026
DA - 2026/11/10
PB - bioRxiv
DO - 10.64898/2026.11.10.765624
UR - https://www.biorxiv.org/content/10.64898/2026.11.10.765624v1
AB - Intrinsically disordered proteins (IDPs) are central to many biological processes, but their function depends on dynamic conformational ensembles rather than a single structure. Despite the impact of artificial intelligence on protein structure prediction, determining these ensembles at atomistic resolution remains challenging. Recently, we introduced bAIes, a framework that leverages the structural information encoded in AlphaFold2 distograms to generate conformational ensembles of IDPs at atomistic resolution. Here, we extend this framework to six state-of-the-art structure prediction tools (AlphaFold2, AlphaFold3, Boltz-1, Boltz-2, ESMFold, and ESMFold2) and systematically compare the structural information encoded in their predicted inter-residue distance distributions. Despite differences between predictors, we show that bAIes generates ensembles with comparable accuracy against experimental NMR and SAXS data, and competitive performance relative to state-of-the-art all-atom simulations and alternative IDP ensemble generation methods. These results demonstrate that bAIes is robust to the choice of structure predictor and that the distance distributions generated by different deep learning architectures contain valuable information about the conformational preferences of disordered proteins. More broadly, our work highlights the potential of combining AI-derived structural information with physics-based modeling to generate atomistic ensembles of intrinsically disordered proteins and regions.
ER -