Accurate enzyme specificity constant prediction with iESC.

Area of Science:

• Biochemistry
• Computational Biology
• Enzyme Kinetics

Background:

• Enzyme specificity constants (ESC), including Michaelis constant (Km) and turnover number (kcat), are crucial for understanding enzyme function.
• Traditional methods for determining Km and kcat are resource-intensive and time-consuming, hindering rapid enzyme characterization.

Purpose of the Study:

• To develop and validate iESC, a novel deep learning model for accurate prediction of enzyme kinetic parameters (Km, kcat, and kcat/Km) using only enzyme sequences and substrate structures.
• To demonstrate iESC's superior performance compared to existing state-of-the-art models.

Main Methods:

• iESC was trained on a large dataset of 41,907 enzyme-substrate kinetic parameters, meticulously curated and preprocessed for accuracy and independence.
• The model integrates advanced feature extraction techniques with deep learning architectures.
• Performance was evaluated using coefficient of determination (R²), root mean squared error (RMSE), and mean absolute error (MAE), and benchmarked against other models.

Main Results:

• iESC achieved high predictive accuracy, with R² values of 0.63 for Km, 0.60 for kcat, and 0.62 for kcat/Km.
• Benchmark tests confirmed iESC's significant outperformance of existing models, showing higher R² and lower RMSE and MAE.
• The model demonstrated strong applicability in high-throughput screening (HTS) and deep mutational scanning (DMS) contexts.

Conclusions:

• iESC offers a rapid, accurate, and sequence-based method for predicting key enzyme kinetic parameters.
• The model has significant implications for accelerating enzyme discovery, engineering, and functional characterization in various biotechnological applications.