Large-scale prospective evaluation of co-folding across hundreds of Mac1–ligand complexes and three virtual screens
8 affiliations
- 1Department of Pharmaceutical Chemistry, University of California San Francisco, San Francisco, CA 94158, USA
- 2Department of Bioengineering and Therapeutic Sciences, University of California San Francisco, San Francisco, CA 94158, USA
- 3Small Molecule Discovery Center, University of California San Francisco, San Francisco, CA 94158, USA
- 4Helen Diller Family Comprehensive Cancer Center, University of California San Francisco, San Francisco, CA 94158, USA
- 5Department of Biological and Medical Informatics, University of California San Francisco, San Francisco, CA 94158, USA
- 6Quantitative Biosciences Institute (QBI), University of California San Francisco, San Francisco, CA 94158, USA
- 7Data Science and Biotechnology Institute, Gladstone Institutes, San Francisco, CA, USA
- 8Department of Cellular and Molecular Pharmacology, University of California San Francisco, San Francisco, CA 94158, USA
Abstract
Accurate prediction of ligand-bound protein complexes and ranking them by affinity are central problems in drug discovery. While deep learning co-folding methods can help address these challenges, their evaluation has been hampered by the difficulties in assessing independence from training data and insufficiently large test sets. Here we test the ability of co-folding methods to predict the structures of 551 ligands, 489 of which are of sufficient quality for evaluation, bound to the SARS-CoV-2 NSP3 macrodomain (Mac1) that were determined after the training cut-off dates. AlphaFold3 (AF3), Boltz-2, and Chai-1 each reproduced >50% of the Mac1 ligand poses to better than 2 Å RMSD of experiment. Despite the potential for co-folding to describe protein conformational changes that stabilize ligand binding, we did not find that common conformational rearrangements, including peptide flip and a large loop opening, were recapitulated by the co-folding prediction. For AF3 and Chai-1, ligand pose prediction confidence weakly, but significantly, tracked experimental potency, while DOCK3.7 energies were only weakly correlated. Boltz-2 affinity predictions showed the strongest correlation with measured potency and, after calibration, achieved lower mean absolute error than a baseline predictor. We next assessed whether co-folding scores could rescore docking hit-lists to distinguish true ligands from non-binders among hundreds of molecules prospectively experimentally tested against AmpC β-lactamase, the dopamine D4 and the σ 2 receptors. AF3 ligand pose confidence values did not separate true ligands from high-scoring false-positives as effectively as docking scores or Boltz-2 affinity predictions did. Taken together, the modest, but independent correlations of docking score and co-folding confidence or affinity suggests that integrating physics-based and deep-learning and approaches may help with hit prioritization and subsequent optimization in structure-based ligand discovery.
Related
- Per-Stage Controls and Failure Modes in a Co-Folding Virtual Screening Cascade for the Keap1–Nrf2 Interaction
- Water-Soluble Diiron Bis-Aminocarbyne Complexes: Photoinduced CO Substitution, Generation of Diiron(I) Aqua-Carbonyl Complexes, and Biological Evaluation
- Targeting redox homeostasis in breast cancer: a review of synthesis, functional mechanisms, and in silico evaluation of metal-based benzimidazole and quinoline complexes
- Virtual screening, synthesis, solid state characterization, evaluation for solubility, dissolution, and stability of gallic acid-based cocrystals of glimepiride
- FROST: certified Hartree–Fock limits of helium and beryllium to hundreds of significant figures
BibTeX
@article{kim2026largescale,
title = {{Large-scale prospective evaluation of co-folding across hundreds of Mac1–ligand complexes and three virtual screens}},
author = {Jongbin Kim and G.J. Correy and Brendan W. Hall and Moira M. Rachman and Olivier Mailhot and Takaya Togo and Ryan L. Gonciarz and Priyadarshini Jaishankar and R. Jeffrey Neitz and Eric R. Hantz and Yagmur Umay Doruk and Maisie GV Stevens and Morgan E. Diolaiti and Rashad L. Reid and Saumya Gopalkrishnan and Nevan J. Krogan and Adam R. Renslo and Alan Ashworth and Brian K. Shoichet and James S. Fraser},
journal = {eLife},
year = {2026},
month = oct,
doi = {10.7554/elife.110475.3},
url = {https://doi.org/10.7554/elife.110475.3},
note = {CC BY},
}
RIS
TY - JOUR
TI - Large-scale prospective evaluation of co-folding across hundreds of Mac1–ligand complexes and three virtual screens
AU - Jongbin Kim
AU - G.J. Correy
AU - Brendan W. Hall
AU - Moira M. Rachman
AU - Olivier Mailhot
AU - Takaya Togo
AU - Ryan L. Gonciarz
AU - Priyadarshini Jaishankar
AU - R. Jeffrey Neitz
AU - Eric R. Hantz
AU - Yagmur Umay Doruk
AU - Maisie GV Stevens
AU - Morgan E. Diolaiti
AU - Rashad L. Reid
AU - Saumya Gopalkrishnan
AU - Nevan J. Krogan
AU - Adam R. Renslo
AU - Alan Ashworth
AU - Brian K. Shoichet
AU - James S. Fraser
PY - 2026
DA - 2026/10/12
JO - eLife
DO - 10.7554/elife.110475.3
UR - https://doi.org/10.7554/elife.110475.3
AB - Accurate prediction of ligand-bound protein complexes and ranking them by affinity are central problems in drug discovery. While deep learning co-folding methods can help address these challenges, their evaluation has been hampered by the difficulties in assessing independence from training data and insufficiently large test sets. Here we test the ability of co-folding methods to predict the structures of 551 ligands, 489 of which are of sufficient quality for evaluation, bound to the SARS-CoV-2 NSP3 macrodomain (Mac1) that were determined after the training cut-off dates. AlphaFold3 (AF3), Boltz-2, and Chai-1 each reproduced >50% of the Mac1 ligand poses to better than 2 Å RMSD of experiment. Despite the potential for co-folding to describe protein conformational changes that stabilize ligand binding, we did not find that common conformational rearrangements, including peptide flip and a large loop opening, were recapitulated by the co-folding prediction. For AF3 and Chai-1, ligand pose prediction confidence weakly, but significantly, tracked experimental potency, while DOCK3.7 energies were only weakly correlated. Boltz-2 affinity predictions showed the strongest correlation with measured potency and, after calibration, achieved lower mean absolute error than a baseline predictor. We next assessed whether co-folding scores could rescore docking hit-lists to distinguish true ligands from non-binders among hundreds of molecules prospectively experimentally tested against AmpC β-lactamase, the dopamine D4 and the σ 2 receptors. AF3 ligand pose confidence values did not separate true ligands from high-scoring false-positives as effectively as docking scores or Boltz-2 affinity predictions did. Taken together, the modest, but independent correlations of docking score and co-folding confidence or affinity suggests that integrating physics-based and deep-learning and approaches may help with hit prioritization and subsequent optimization in structure-based ligand discovery.
SN - 2050-084X
ER -