Abstract

Enzyme specificity constants (ESC) are critical quantitative metrics of enzyme properties, especially the Michaelis constant ( K m ) and turnover number ( k cat ). However, traditional methods for measuring K m and k cat are laborious and time-consuming. Here, we introduce iESC, a deep learning model that accurately predicts these parameters based solely on enzyme sequences and substrate structures. iESC was developed using a comprehensive dataset of 41,907 enzyme-substrate kinetic parameters compiled from existing databases and reports. Rigorous data preprocessing ensured independence and accuracy. By integrating multiple advanced feature extraction and deep learning techniques, iESC achieved coefficient of determination (R 2 ) values of 0.63, 0.60, and 0.62 for K m , k cat , and k cat / K m , respectively. Benchmark tests demonstrated that iESC significantly outperformed existing state-of-the-art models, with higher R 2 values and lower root mean squared error (RMSE) and mean absolute error (MAE) on various datasets. We further validated iESC’s outstanding applicability in the high-throughput screening (HTS) and deep mutational scanning (DMS) technologies of enzymes.