Deep Learning-Based Prediction of Enzyme Optimal pH and Design of Point Mutations to Improve Acid Resistance

Area of Science:

• Biochemistry
• Computational Biology
• Enzyme Engineering

Background:

• Enzyme activity is significantly influenced by pH, necessitating accurate prediction of optimal pH.
• Understanding pH-dependent enzyme kinetics is crucial for various biotechnological applications.

Purpose of the Study:

• To develop an accurate deep learning predictor for enzyme optimal pH.
• To create a computational tool for identifying enzyme pH preferences and guiding enzyme design.

Main Methods:

• Developed CatOpt, a deep learning model for predicting enzyme optimal pH.
• Utilized residue attention weights for model interpretability.
• Applied CatOpt for classifying acidophilic/alkaliphilic enzymes and predicting pH shifts from mutations.

Main Results:

• CatOpt outperformed existing predictors with RMSE = 0.833 and R² = 0.479.
• The model demonstrated interpretability through residue attention weights.
• Successfully predicted pH shifts and guided a mutation enhancing enzyme activity at low pH by ~7%.

Conclusions:

• CatOpt is an effective computational tool for predicting enzyme optimal pH and pH preferences.
• The model shows promise for *in silico* enzyme design, particularly for pH-dependent activities.