Biomolecular design through latent reasoning with RFD4-Proteina

Biomolecular design requires finding chemically and geometrically compatible arrangements of interacting atoms and molecules. The complexity of problems which can be tackled by deep learning design models is limited by both the molecular landscape they encompass and their ability to explore that landscape effectively. Here we introduce RFD4-Proteina, which addresses these limitations through intrinsic and extrinsic reasoning within a shared chemical and conformational latent space. The latent space encompasses standard and modified amino acids, sugars, DNA, and RNA, with force-field supervision favoring physically plausible atomic reconstructions. Within this space, intrinsic reasoning uses recurrent refinement to adaptively allocate computation, while extrinsic reasoning guides search with arbitrary constraints such as energetic or geometric criteria. Across in silico protein design benchmarks, RFD4-Proteina achieves higher design success rates at greater generation throughput than all previous methods evaluated, with further gains from conformational flexibility, recurrent refinement, and search. The same model performs equivalently to or better than state-of-the-art specialist methods for side-chain packing, sequence design, conformational sampling, and design refinement. We demonstrate the power and versatility of RFD4-Proteina experimentally by designing specific and cross-reactive binders to therapeutically relevant targets, recognizing phosphorylations, controlling glycosylation, and creating enzymes that catalyze both bond-breaking and bond-forming reactions, including a first proof-of-concept demonstration of a de novo designed enzyme for active pharmaceutical ingredient synthesis. By unifying diverse chemistry, conformational flexibility, and scalable reasoning in a single physically grounded latent space, RFD4-Proteina expands the frontier of biomolecular design.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-15
DOI
https://doi.org/10.64898/2026.11.12.766037
Primary Topic
Protein Structure and Dynamics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Biomolecular design through latent reasoning with RFD4-Proteina

Brian Coventry, Edin Muratspahić, Preetham Venkatesh, Michael M. Bronstein et al.
bioRxiv (Cold Spring Harbor Laboratory)
Protein Structure and Dynamics
preprint

Biomolecular design through latent reasoning with RFD4-Proteina

Brian Coventry, Edin Muratspahić, Preetham Venkatesh, Michael M. Bronstein, Donghyo Kim, Alexis Courbet, Neil P. King, Yehlin Cho, Riti Biswas, Rohith Krishna, Helen E. Eisenach, Kieran Didi, Karsten Kreis, Jie Chen, Brianne R. King, Andrew C. Hunt, Arash Vahdat, Margaret C. Lunn-Halbert, Per Jr Greisen, Magnus S. Bauer, Donald Hilvert, Michel Sanchez, Maximilian Gantz, Max Crispin, Gaurav Bhardwaj, David A. Baker, Hojeong Shin, David E. Kim, Shajesh Sharma, Anthony P. Green, Inna Goreshnik, Andrew Favor, Paul T. Kim, Zefei Xu, Nora K. McNamara-Bordewick, Rafael I. Brent, Joel D. Allen, Seth M. Woodbury, Tomas Geffner, Christian Dallago, Lemuel Chang, Adam M. Chazin-Gray, DéJenaé L. See, Oliver Nagl, Yuyang Zhang, Kejia Wu, Henry Jones, Griffin H. Rangel, Frank DiMaio, Nathaniel Corley, Saman Salike, Julia E. Bonzanini, Elliot S. Delfosse, Antonia Ruder, Maxwell W. Kazman, Aiko Muraishi, Yanjing Li, Jasper Butcher, Eric Sun, Yuwei Yang, Zi Ming Li
preprint en

Abstract

Biomolecular design requires finding chemically and geometrically compatible arrangements of interacting atoms and molecules. The complexity of problems which can be tackled by deep learning design models is limited by both the molecular landscape they encompass and their ability to explore that landscape effectively. Here we introduce RFD4-Proteina, which addresses these limitations through intrinsic and extrinsic reasoning within a shared chemical and conformational latent space. The latent space encompasses standard and modified amino acids, sugars, DNA, and RNA, with force-field supervision favoring physically plausible atomic reconstructions. Within this space, intrinsic reasoning uses recurrent refinement to adaptively allocate computation, while extrinsic reasoning guides search with arbitrary constraints such as energetic or geometric criteria. Across in silico protein design benchmarks, RFD4-Proteina achieves higher design success rates at greater generation throughput than all previous methods evaluated, with further gains from conformational flexibility, recurrent refinement, and search. The same model performs equivalently to or better than state-of-the-art specialist methods for side-chain packing, sequence design, conformational sampling, and design refinement. We demonstrate the power and versatility of RFD4-Proteina experimentally by designing specific and cross-reactive binders to therapeutically relevant targets, recognizing phosphorylations, controlling glycosylation, and creating enzymes that catalyze both bond-breaking and bond-forming reactions, including a first proof-of-concept demonstration of a de novo designed enzyme for active pharmaceutical ingredient synthesis. By unifying diverse chemistry, conformational flexibility, and scalable reasoning in a single physically grounded latent space, RFD4-Proteina expands the frontier of biomolecular design.

bioRxiv (Cold Spring Harbor Laboratory)
Howard Hughes Medical Institute (US), Austrian Academy of Sciences (AT), Duke University (US), University of Washington (US), University of Manchester (GB), ETH Zurich (CH), University of Oxford (GB), Nvidia (United States) (US), Max Planck Institute for Terrestrial Microbiology (DE), University of Southampton (GB), AITHYRA GmbH- Research Institute for Biomedical Artificial Intelligence (Austria) (AT), Laboratory of Organic Chemistry (CH), Tsinghua University (CN)
Protein Structure and Dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon