A Multimodal Ensemble Framework for Optimal Mutant Prediction and Computational Enzyme Engineering.

Area of Science:

• Biochemistry
• Protein Engineering
• Computational Biology

Background:

• Enzyme engineering is crucial for developing improved biocatalysts.
• Traditional methods struggle to model complex protein interactions (epistasis and long-range effects).

Purpose of the Study:

• To present GEMS, a novel framework for enzyme engineering.
• To leverage ensemble zero-shot learning across multiple data modalities for predicting beneficial enzyme variants.

Main Methods:

• GEMS integrates evolutionary, structural, and sequence data.
• It models the sequence-structure-function relationship using ensemble zero-shot capabilities.
• Benchmarking against state-of-the-art methods was performed.

Main Results:

• GEMS demonstrates competitive performance in ranking beneficial variants.
• It excels at capturing long-range functional constraints, generating informative variant libraries.
• Applied to five diverse enzymes, GEMS identified mutations improving catalytic efficiency by 1.1 to 3.2-fold.

Conclusions:

• GEMS is a powerful and versatile tool for advanced enzyme engineering.
• The framework effectively models complex protein interactions for improved enzyme function.