Skip to content
DraconiticResearch
Search Sources API
203,234 studies31 sourcesUpdated 13 min ago
Peer-reviewedCC BYNature Communications

FLOWR.ROOT – A flow matching-based foundation model for joint multi-purpose structure-aware 3D ligand generation and affinity prediction

  • Julian Cremer1,2
  • Tuan Le2
  • Mohammad Mehdi Ghahremanpour3,4
  • Emilia Sługocka5,6
  • Filipe Menezes7,8,9,10
  • Djork-Arné Clevert2
10 affiliations
  1. 1Machine Learning & Computational Sciences, Pfizer Worldwide R&D, Berlin, Germany. [email protected]
  2. 2Machine Learning & Computational Sciences, Pfizer Worldwide R&D, Berlin, Germany
  3. 3Computational Chemistry, Medicine Design, Pfizer Worldwide R&D, Cambridge, MA, USA
  4. 4Computational Chemistry, Medicine Design, Pfizer Worldwide R&D, Cambridge, USA
  5. 5Department of Physicochemical Drug Analysis, Faculty of Pharmacy, Jagiellonian University Medical College, Cracow, Poland
  6. 6Doctoral School of Medical and Health Sciences, Jagiellonian University Medical College, Cracow, Poland
  7. 7Institute of Structural Biology, Molecular Targets and Therapeutics Center, Helmholtz Munich, Neuherberg, Germany. [email protected]
  8. 8TUM School of Natural Sciences, Department of Bioscience, Bayerisches NMR Zentrum, Technical University of Munich, Garching, Germany. [email protected]
  9. 9Institute of Structural Biology, Molecular Targets and Therapeutics Center, Helmholtz Munich, Neuherberg, Germany
  10. 10TUMSchool of Natural Sciences, Department of Bioscience, Bayerisches NMR Zentrum, Technical University of Munich, Garching, Germany
Open at sourcePDFDOICite

Abstract

We present FLOWR.ROOT, an S E (3)-equivariant flow-matching foundation model that unifies pocket-aware 3D ligand generation with multi-endpoint binding affinity prediction (pIC 50 , p K i , p K d , pEC 50 ) and pLDDT-based confidence estimation in a single backbone. One trained model supports de novo pocket-conditional generation, interaction- and pharmacophore-conditional sampling, scaffold hopping and elaboration, and fragment growing or replacement, enabled by a mixed isotropic–anisotropic prior placement strategy. Training proceeds in three stages: large-scale pre-training on billions of ligand conformations and millions of mixed-fidelity protein–ligand complexes, refinement on curated co-crystal data, and project-specific adaptation via parameter-efficient LoRA finetuning. Joint structure–affinity modelling enables inference-time importance-sampling guidance for single- and multi-objective design without external scoring functions. Case studies on kinase selectivity (CK2 α /CLK3) and scaffold elaboration on TYK2, ER α , and BACE1 illustrate utility from hit identification through lead optimization.

Related

  1. Beyond the Ligand Applicability Domain: Generalization and Local Resolution in Structure-Aware Foundation Models for Affinity Prediction
    Pieter B. Burger, Thomas M. Kaiser, Parker Dryja +7
    ChemRxivPreprintCC BYFull textGeneral chemistryChemRxiv
  2. HierDTA: Hierarchical Structure-Aware Representation Learning for Drug–Target Affinity Prediction
    Huangchao Xu, Baohua Zhang, Qian Liu +2
    ACS OmegaPeer-reviewedCC BY-NC-ND 4.0Full textGeneral chemistryCrossrefPMC
  3. Reliability of AI Methods in Drug Discovery: Evaluation of Boltz-2 for Structure and Binding Affinity Prediction
    Shunzhou Wan, Xibei Zhang, Xue Xiao +1
    Journal of Chemical Theory and ComputationPeer-reviewedCC BYFull text20 citedMedicinal chemistry & drug discoveryOpenAlex
  4. FLOWR: flow matching for structure-aware de novo, interaction- and fragment-based ligand generation
    Julian Cremer, Ross Irwin, Alessandro Tibo +3
    Nature Computational SciencePeer-reviewedCC BYFull text24 citedMedicinal chemistry & drug discoveryOpenAlex
  5. Curvature-Aware Flow Matching for Molecular Structure Generation
    Samir Darouich, Juliane Geisler, Jacob W. Toney +4
    arXivPreprintCC BY 4.0Full textPhysical chemistryarXiv

Cite

BibTeXRIS
BibTeX
@article{cremer2026flowrroot,
  title = {{FLOWR.ROOT – A flow matching-based foundation model for joint multi-purpose structure-aware 3D ligand generation and affinity prediction}},
  author = {Julian Cremer and Tuan Le and Mohammad Mehdi Ghahremanpour and Emilia Sługocka and Filipe Menezes and Djork-Arné Clevert},
  journal = {Nature Communications},
  year = {2026},
  month = jul,
  doi = {10.1038/s41467-026-74130-9},
  url = {https://doi.org/10.1038/s41467-026-74130-9},
  note = {CC BY},
}
RIS
TY  - JOUR
TI  - FLOWR.ROOT – A flow matching-based foundation model for joint multi-purpose structure-aware 3D ligand generation and affinity prediction
AU  - Julian Cremer
AU  - Tuan Le
AU  - Mohammad Mehdi Ghahremanpour
AU  - Emilia Sługocka
AU  - Filipe Menezes
AU  - Djork-Arné Clevert
PY  - 2026
DA  - 2026/07/06
JO  - Nature Communications
DO  - 10.1038/s41467-026-74130-9
UR  - https://doi.org/10.1038/s41467-026-74130-9
AB  - We present FLOWR.ROOT, an S E (3)-equivariant flow-matching foundation model that unifies pocket-aware 3D ligand generation with multi-endpoint binding affinity prediction (pIC 50 , p K i , p K d , pEC 50 ) and pLDDT-based confidence estimation in a single backbone. One trained model supports de novo pocket-conditional generation, interaction- and pharmacophore-conditional sampling, scaffold hopping and elaboration, and fragment growing or replacement, enabled by a mixed isotropic–anisotropic prior placement strategy. Training proceeds in three stages: large-scale pre-training on billions of ligand conformations and millions of mixed-fidelity protein–ligand complexes, refinement on curated co-crystal data, and project-specific adaptation via parameter-efficient LoRA finetuning. Joint structure–affinity modelling enables inference-time importance-sampling guidance for single- and multi-objective design without external scoring functions. Case studies on kinase selectivity (CK2 α /CLK3) and scaffold elaboration on TYK2, ER α , and BACE1 illustrate utility from hit identification through lead optimization.
SN  - 2041-1723
ER  -