Latent generative search unlocks de novo design of untapped biomolecular interactions at scale
- 1NVIDIA Corporation, Santa Clara, CA, USA
Abstract
De novo protein design has advanced rapidly, yet designing binders to polar, solvent-exposed epitopes and small, flexible ligands remains challenging. Such hydrated surfaces and flexible molecules, including carbohydrates, provide few of the hydrophobic contacts favoured by current methods and have largely resisted de novo binders. To address this challenge, here we introduce latent generative search for binder design, a novel framework that uses reward-guided search at inference time to steer the Proteina-Complexa generative model. The model codesigns sequence and structure - generating them together in a continuous latent space - and thereby removes the inverse-folding step on which current methods rely. In a screen of more than one million designs by multiplexed phage display, latent generative search produced more validated binders than every other method tested, its codesigned sequences surpassing post hoc redesign. It delivered high-affinity binders across therapeutic receptors, a viral attachment protein and intracellular signalling targets. Our approach also accessed previously untapped biology, generating the first de novo proteins that bind a free carbohydrate, including one that discriminates between blood-group antigens - a polar, flexible target class beyond the reach of current design methods.
Related
- Inferring the latent network of pairwise mutualistic preferences from observed plant-pollinator interactions
- Global Sesame Germplasm Harbors Untapped Genetic Potential for Crop Improvement
- SARDINE unlocks EWAS summary statistics for large-scale epigenetic biomarker analysis
- Surface immune signaling unlocks NLR activation through mRNA alternative splicing
- A NEST-based framework unlocks massively parallel simulation of networks of multi-compartment neurons with customizable subcellular dynamics
BibTeX
@misc{didi2026latent,
title = {{Latent generative search unlocks de novo design of untapped biomolecular interactions at scale}},
author = {K. Didi and D. Reidenbach and M. Penner and S. Ravichandran and M. Case and M. Nichols and E. Swanson and A. Reis and M. Prescott and Y. Qian and D. Qian and J. Yang and W. Li and L. Li and D. Shonai and S. Gay and B. Basu Mallik and H. Y. Chim and L. Chen and M. Atienza Juanatey and H. Klein and D. Rieger and P. Schlegel and A. U. Macintyre and M. Secor and D. Granata and S. Cha and Z. Cao and G. Zhou and T. Geffner and X. Chen and M. Livne and Z. Zhang and T. Zhang and K. Gion and M. M. Bronstein and M. Steinegger and K. Deibler and S. Soderling and C. T. Schoeder and A. Khmelinskaia and F. Hollfelder and C. Dallago and E. Kucukbenli and A. Vahdat and P. Ogden and K. Kreis},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.12.751118},
url = {https://www.biorxiv.org/content/10.64898/2026.09.12.751118v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Latent generative search unlocks de novo design of untapped biomolecular interactions at scale
AU - K. Didi
AU - D. Reidenbach
AU - M. Penner
AU - S. Ravichandran
AU - M. Case
AU - M. Nichols
AU - E. Swanson
AU - A. Reis
AU - M. Prescott
AU - Y. Qian
AU - D. Qian
AU - J. Yang
AU - W. Li
AU - L. Li
AU - D. Shonai
AU - S. Gay
AU - B. Basu Mallik
AU - H. Y. Chim
AU - L. Chen
AU - M. Atienza Juanatey
AU - H. Klein
AU - D. Rieger
AU - P. Schlegel
AU - A. U. Macintyre
AU - M. Secor
AU - D. Granata
AU - S. Cha
AU - Z. Cao
AU - G. Zhou
AU - T. Geffner
AU - X. Chen
AU - M. Livne
AU - Z. Zhang
AU - T. Zhang
AU - K. Gion
AU - M. M. Bronstein
AU - M. Steinegger
AU - K. Deibler
AU - S. Soderling
AU - C. T. Schoeder
AU - A. Khmelinskaia
AU - F. Hollfelder
AU - C. Dallago
AU - E. Kucukbenli
AU - A. Vahdat
AU - P. Ogden
AU - K. Kreis
PY - 2026
DA - 2026/09/12
PB - bioRxiv
DO - 10.64898/2026.09.12.751118
UR - https://www.biorxiv.org/content/10.64898/2026.09.12.751118v1
AB - De novo protein design has advanced rapidly, yet designing binders to polar, solvent-exposed epitopes and small, flexible ligands remains challenging. Such hydrated surfaces and flexible molecules, including carbohydrates, provide few of the hydrophobic contacts favoured by current methods and have largely resisted de novo binders. To address this challenge, here we introduce latent generative search for binder design, a novel framework that uses reward-guided search at inference time to steer the Proteina-Complexa generative model. The model codesigns sequence and structure - generating them together in a continuous latent space - and thereby removes the inverse-folding step on which current methods rely. In a screen of more than one million designs by multiplexed phage display, latent generative search produced more validated binders than every other method tested, its codesigned sequences surpassing post hoc redesign. It delivered high-affinity binders across therapeutic receptors, a viral attachment protein and intracellular signalling targets. Our approach also accessed previously untapped biology, generating the first de novo proteins that bind a free carbohydrate, including one that discriminates between blood-group antigens - a polar, flexible target class beyond the reach of current design methods.
ER -