{
  "id": 8288610,
  "title": "Physical priors improve performance of structure-based binding affinity models",
  "url": "https://urgent.news/2026/09/18/physical-priors-improve-performance-of-structure-based-binding",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-18T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.09.11.750982v1?rss=1"
  },
  "original_language": "en",
  "account": "Structure-based drug discovery utilizes structural information to design novel therapeutic small molecules. However, these models have not yet surpassed 2D ligand-only models in machine learning. Researchers have found that physics-based priors can enhance the performance of structure-based models. By comparing various model architectures and physical priors on different prediction tasks, they demonstrated that incorporating roto-translational inductive biases through E(3)-invariant and E(3)-equivariant architectures, along with optimized methods for combining learned embeddings, improved predictive performance. This modular training and evaluation of neural networks (mtenn) package optimized several aspects of model construction using the PDBBind dataset, resulting in improved performance and data efficiency. When applied to the COVID Moonshot small molecule drug discovery dataset, the tuned models matched the accuracy of industry-standard ligand-only models. The study highlights that encoding physical priors can enhance model performance, while more complex biases like equivariance offer limited benefits. Moreover, structure-based models exhibit better generalization to unseen targets and higher training efficiency. These findings emphasize the importance of incorporating physics-informed constraints in model architecture development, particularly for tasks focused on generalizability, providing guidance for their application in early-stage drug discovery campaigns.",
  "summary": "Structure-based drug discovery is a widely used paradigm for the rational design of novel small molecule therapeutics. However, the benefits conferred by the use of structural information has seen limited adoption in machine learning, where ligand-only (\"2D\") models are still the industry standard for molecular property or binding affinity prediction. Structure-based (\"3D\") ML models for…",
  "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."
}