EnzyACT: A Novel Deep Learning Method to Predict the Impacts of Single and Multiple Mutations on Enzyme Activity

Area of Science:

• Biotechnology
• Computational Biology
• Biochemistry

Background:

• Enzyme engineering aims to tailor enzymes for specific applications via mutations.
• A key challenge is the activity-stability trade-off, where improving one often impairs the other.
• Predicting mutation impacts on enzyme stability is established, but predicting activity changes remains difficult.

Purpose of the Study:

• To develop a fast and accurate method for predicting mutation-induced changes in enzyme activity.
• To aid in enzyme design and deepen the understanding of the enzyme activity-stability trade-off.
• To provide insights into the role of distant mutations in enzyme activity modulation.

Main Methods:

• Introduced EnzyACT, a novel deep learning approach.
• Fused graph techniques with protein embedding for activity prediction.
• Trained the model on a curated dataset of single and multiple-point mutations.

Main Results:

• EnzyACT demonstrated uniform performance across multiple independent benchmark datasets.
• The model effectively predicts activity changes for both single and multiple mutations.
• Analysis provided insights into how distant mutations influence enzyme activity.

Conclusions:

• EnzyACT offers a valuable tool for predicting enzyme activity changes, facilitating enzyme engineering.
• The method aids in understanding and potentially overcoming the activity-stability trade-off.
• This work contributes to improved enzyme design strategies and the identification of key catalytic residues.