Model-Derived Mechanistic Insights into Structural, Surface, and Physicochemical Determinants Underlying Enzyme pH...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Cheminformatics
• Enzyme Engineering and Industrial Biotechnology

Background:

• Optimal pH is critical for enzyme catalytic efficiency and stability, impacting industrial biocatalysis.
• Traditional experimental methods for determining optimal pH are time-consuming and costly.
• Existing computational models struggle with comprehensive feature representation for accurate pH prediction.

Purpose of the Study:

• To propose a novel multi-source attention Kolmogorov-Arnold Network (EnzMSA-KAN) for predicting enzyme optimal pH.
• To develop a computational tool that overcomes the limitations of traditional experimental and existing modeling approaches.
• To enhance the efficiency and accuracy of enzyme function prediction for industrial applications.

Main Methods:

• Developed a three-stage EnzMSA-KAN model integrating protein sequence evolutionary information, residue physicochemical properties, and global statistical features.
• Employed a two-layer Graph Attention Network with multi-head attention pooling to capture spatial and sequential residue dependencies and focus on key functional regions.
• Utilized a regularized Kolmogorov-Arnold Network (KAN) regression head for accurate fitting of complex non-linear relationships between enzyme microenvironment and optimal pH.

Main Results:

• EnzMSA-KAN achieved a Mean Absolute Error (MAE) of 0.57, Root Mean Square Error (RMSE) of 0.81, and R-squared (R²) of 0.52 on a benchmark dataset.
• The model significantly outperformed existing mainstream methods in predicting enzyme optimal pH.
• Attention visualization confirmed the model's ability to identify key functional residues, providing biological interpretability.

Conclusions:

• EnzMSA-KAN offers an efficient and accurate computational tool for predicting enzyme optimal pH.
• The model's deep feature fusion and attention mechanisms enable effective capture of crucial enzyme characteristics.
• This research provides a new computational paradigm for the rational design and engineering of industrial enzymes.