{
  "id": 13978986,
  "title": "Large-scale prospective evaluation of co-folding across hundreds of Mac1–ligand complexes and three virtual screens",
  "url": "https://urgent.news/2026/10/12/large-scale-prospective-evaluation-of-co-folding-across-hundreds-of",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-12T00:00:00.000Z",
  "source": {
    "name": "eLife",
    "slug": "elife",
    "url": "https://elifesciences.org/articles/110475"
  },
  "original_language": "en",
  "account": "Predicting the structures of protein-ligand complexes and ranking them by affinity is a crucial challenge in drug discovery. Deep-learning co-folding techniques can assist, but evaluating their performance is hindered by difficulties in ensuring independence from training data and insufficiently large test sets. In this study, the ability of co-folding methods to predict the structures of 551 ligands, 489 of which were of high quality and bound to the SARS-CoV-2 NSP3 macrodomain, were assessed. AlphaFold3 (AF3), Boltz-2, and Chai-1 each reproduced 50% of the ligand poses with a root-mean-square deviation of less than 2 Å from experimental data. Although co-folding can potentially describe conformational changes stabilizing ligand binding, common rearrangements like peptide flip and large loop opening were not accurately captured by co-folding predictions. For AF3 and Chai-1, the confidence in ligand pose prediction weakly and significantly tracked experimental potency, while DOCK3.7 energies showed a weak correlation. Boltz-2 affinity predictions demonstrated the strongest correlation with measured potency, and after calibration, achieved a lower mean absolute error than a baseline predictor. Lastly, the study explored whether co-folding scores could rescore docking hit lists to differentiate true ligands from non-binders among hundreds of experimentally tested molecules against AmpC β-lactamase, dopamine D4, and σ2 receptors. While AF3 ligand pose confidence values were not as effective as docking scores or Boltz-2 affinity predictions in separating true ligands from high-scoring false positives, the modest yet independent correlations between docking scores and co-folding confidence or affinity suggest that combining physics-based and deep-learning approaches could aid hit prioritization and subsequent optimization in structure-based ligand discovery.",
  "summary": "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…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}