EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Dual-encoder contrastive learning accelerates enzyme discovery

Jason W Rocks iD, Dat P Truong iD, Dmitrij Rappoport, Samuel Maddrell-Mander, Daniel A Martin-Alarcon iD, Toni M Lee iD, Steven Crossan, Joshua E Goldford iD

Proceedings of the National Academy of Sciences · 24.03.2026 · 123 (12)

DOI: 10.1073/pnas.2520070123 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the National Academy of Sciences, online erschienen am 17.03.2026. Verfasst von Jason W Rocks und 7 weiteren Autor:innen. Beteiligte Einrichtung: Dayhoff Labs, Inc., Cambridge, MA 02140. Hauptthemen laut MEDLINE: Enzymes, Deep Learning, Computational Biology. Die Arbeit wurde bisher 9-mal zitiert (Crossref) und nennt 58 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: U.S. Department of Energy (DOE).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The ability to engineer enzymes for desired reactions is a cornerstone of modern biotechnology, yet identifying suitable starting proteins remains a critical bottleneck. Although contrastive learning offers a compelling computational approach for enzyme discovery, these models have yet to be implemented at scale or proven effective in real-world experimental settings. Here, we present Horizyn-1 , a computationally efficient deep learning framework that enables large-scale reaction-to-enzyme recommendation validated through comprehensive experimental testing. Leveraging a combination of reaction fingerprints and protein language models, we trained Horizyn-1 on millions of reaction–enzyme pairs to achieve state-of-the-art performance, recovering an enzyme with correct activity within the top 100 hits for over 75% of test reactions. We experimentally validate Horizyn-1 across three enzyme discovery scenarios: identifying enzymes for orphan reactions, predicting enzyme promiscuity for both characterized and uncharacterized enzymes, and discovering enzymes for nonnatural biochemical reactions including lysine-driven transaminations that enable efficient synthesis of noncanonical amino acids. On underrepresented reaction classes, we find that fine-tuning with fewer than 10 additional reactions can dramatically improve performance. Furthermore, a logarithmic scaling of model performance with training dataset size suggests continued improvement with larger and more diverse reaction datasets. Horizyn-1 addresses the critical bottleneck of sourcing initial enzymes for optimization campaigns, enabling efficient and scalable in silico screening for enzymes with desired activities and promising to accelerate future efforts in biocatalysis and metabolic engineering.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

EnzymesDeep LearningComputational BiologyProtein Engineering

Stichwörter der Autor:innen

AIdeep learningenzyme discoveryenzymology

Substanzen

Enzymes

Autor:innen und Institutionen

  • Dayhoff Labs, Inc., Cambridge, MA 02140.

ORCID-Profile: Jason W Rocks, Dat P Truong, Daniel A Martin-Alarcon, Toni M Lee, Joshua E Goldford

Förderung

  • U.S. Department of Energy (DOE)

Zitieren

APA

Rocks, J. W., Truong, D. P., Rappoport, D., Maddrell-Mander, S., Martin-Alarcon, D. A., Lee, T. M., Crossan, S., & Goldford, J. E. (2026). Dual-encoder contrastive learning accelerates enzyme discovery. Proceedings of the National Academy of Sciences, 123(12). https://doi.org/10.1073/pnas.2520070123

Vancouver

Rocks JW, Truong DP, Rappoport D, Maddrell-Mander S, Martin-Alarcon DA, Lee TM, et al. Dual-encoder contrastive learning accelerates enzyme discovery. Proceedings of the National Academy of Sciences. 2026;123(12). doi:10.1073/pnas.2520070123

BibTeX
@article{Rocks2026,
  author = {Rocks, Jason W and Truong, Dat P and Rappoport, Dmitrij and Maddrell-Mander, Samuel and Martin-Alarcon, Daniel A and Lee, Toni M and Crossan, Steven and Goldford, Joshua E},
  title = {Dual-encoder contrastive learning accelerates enzyme discovery},
  journal = {Proceedings of the National Academy of Sciences},
  year = {2026},
  volume = {123},
  number = {12},
  doi = {10.1073/pnas.2520070123}
}
Kostenlos starten →