Pearl: A Foundation Model for Placing Every Atom in the Right Location

  • 2025-10-28 17:36:51
  • Genesis Research Team, Alejandro Dobles, Nina Jovic, Kenneth Leidal, Pranav Murugan, David C. Williams, Drausin Wulsin, Nate Gruver, Christina X. Ji, Korrawat Pruegsanusak, Gianluca Scarpellini, Ansh Sharma, Wojciech Swiderski, Andrea Bootsma, Richard Strong Bowen, Charlotte Chen, Jamin Chen, Marc André Dämgen, Roy Tal Dew, Benjamin DiFrancesco, J. D. Fishman, Alla Ivanova, Zach Kagin, David Li-Bland, Zuli Liu, Igor Morozov, Jeffrey Ouyang-Zhang, Frank C. Pickard IV, Kushal S. Shah, Ben Shor, Gabriel Monteiro da Silva, Maxx Tessmer, Carl Tilbury, Cyr Vetcher, Daniel Zeng, Maruan Al-Shedivat, Aleksandra Faust, Evan N. Feinberg, Michael V. LeVine, Matteus Pan
  • 0

Abstract

Accurately predicting the three-dimensional structures of protein-ligandcomplexes remains a fundamental challenge in computational drug discovery thatlimits the pace and success of therapeutic design. Deep learning methods haverecently shown strong potential as structural prediction tools, achievingpromising accuracy across diverse biomolecular systems. However, theirperformance and utility are constrained by scarce experimental data,inefficient architectures, physically invalid poses, and the limited ability toexploit auxiliary information available at inference. To address these issues,we introduce Pearl (Placing Every Atom in the Right Location), a foundationmodel for protein-ligand cofolding at scale. Pearl addresses these challengeswith three key innovations: (1) training recipes that include large-scalesynthetic data to overcome data scarcity; (2) architectures that incorporate anSO(3)-equivariant diffusion module to inherently respect 3D rotationalsymmetries, improving generalization and sample efficiency, and (3)controllable inference, including a generalized multi-chain templating systemsupporting both protein and non-polymeric components as well as dualunconditional/conditional modes. Pearl establishes a new state-of-the-artperformance in protein-ligand cofolding. On the key metric of generatingaccurate (RMSD < 2 \r{A}) and physically valid poses, Pearl surpasses AlphaFold3 and other open source baselines on the public Runs N' Poses and PoseBustersbenchmarks, delivering 14.5% and 14.2% improvements, respectively, over thenext best model. In the pocket-conditional cofolding regime, Pearl delivers$3.6\times$ improvement on a proprietary set of challenging, real-world drugtargets at the more rigorous RMSD < 1 \r{A} threshold. Finally, we demonstratethat model performance correlates directly with synthetic dataset size used intraining.

 

Quick Read (beta)

loading the full paper ...