A geometric foundation model for enzyme retrieval with evolutionary insights

2026-02-12 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Engineering & Applied Sciences · Depth: Expert, long

Summary

EnzymeCAGE, a catalytic-specific geometric foundation model, has been introduced to address the complexities of enzyme-reaction relationships. Trained on approximately 1.5 million structure-informed enzyme–reaction pairs from over 3,000 species, EnzymeCAGE integrates a geometry-aware multimodal architecture with evolutionary information. This model effectively maps dependencies between enzyme structure, catalytic function, and reaction specificity. It accommodates both experimental and predicted enzyme structures and is applicable across diverse enzyme families and metabolites. Extensive evaluations demonstrate EnzymeCAGE's state-of-the-art performance in enzyme function prediction, reaction de-orphaning, catalytic site identification, and biosynthetic pathway reconstruction, indicating its potential to accelerate biocatalyst discovery and engineering. The model's source code is available on GitHub, and its training data utilizes public databases like Rhea and AlphaFold.

Key takeaway

For AI Researchers and Computational Biologists working on enzyme engineering, EnzymeCAGE offers a robust tool for accelerating biocatalyst discovery. Your team should explore integrating this open-source model into workflows for enzyme function prediction and pathway reconstruction, leveraging its state-of-the-art performance to enhance the design and optimization of novel enzymes. Consider its applicability to both experimental and AlphaFold-predicted structures for broader utility.

Key insights

EnzymeCAGE is a geometric foundation model that predicts enzyme function and reaction specificity using structural and evolutionary data.

Principles

• Integrate geometry-aware multimodal architecture.
• Incorporate evolutionary information for specificity.
• Utilize large datasets of enzyme-reaction pairs.

Method

EnzymeCAGE models enzyme structure, catalytic function, and reaction specificity by integrating a geometry-aware multimodal architecture with evolutionary information, trained on 1.5 million enzyme-reaction pairs.

In practice

• Predict enzyme function for novel enzymes.
• Identify catalytic sites in enzyme structures.
• Reconstruct biosynthetic pathways.

Topics

• EnzymeCAGE
• Geometric Foundation Models
• Enzyme Function Prediction
• Biocatalyst Engineering
• Evolutionary AI

Code references

• samgoldman97/enzyme-datasets
• GENTEL-lab/EnzymeCAGE

Best for: AI Researcher, AI Scientist, Research Scientist

Related on AIssential

• Identification and engineering of highly functional potyviral proteases in cells using co-evolutionary models — Co-evolutionary models can accurately predict and engineer superior potyviral proteases for precise cellular control.
• EnzymeTuning improves enzyme-constrained metabolic modeling and proteome abundance prediction through deep learning — EnzymeTuning, a GAN-based framework, optimizes *k*cat values to enhance metabolic modeling and proteome prediction accuracy.
• Two-dimensional geometric template diffusion for boosting single-sequence protein structure prediction — TDFold uses 2D geometric template diffusion for efficient, accurate single-sequence protein structure prediction.
• Imperagen raises £5M seed round to accelerate AI-driven enzyme engineering — Imperagen's closed-loop AI platform integrates quantum simulation and automated wet lab feedback to rapidly engineer enzymes, significantly boosting productivity.
• 🔬Beyond AlphaFold: How Boltz is Open-Sourcing the Future of Drug Discovery — AlphaFold's legacy drives open-source tools for democratizing advanced molecular structure prediction and generative design.
• Bridging three-dimensional molecular structures and artificial intelligence with a conformation description language — ConfSeq enables large language models to process 3D molecular structures by converting them into discrete token sequences.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.