Hydroxylase Thermostability Prediction Based on Self-Trained Semisupervised Iteration and Bayesian Dynamic Tuning.

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Bioinformatics
• Enzyme Engineering

Background:

• Existing enzyme thermostability prediction models often lack specificity for hydroxylases, limiting their application in targeted enzyme design.
• Hydroxylase-specific models are crucial for improving accuracy and efficiency in protein engineering efforts.
• Developing accurate predictive tools for hydroxylase thermostability is essential for reducing experimental costs and time.

Purpose of the Study:

• To develop a specialized, self-trained semisupervised framework (HyS-BST) for accurate hydroxylase thermostability prediction.
• To improve the prediction of mutant thermostability in terms of ΔΔG for hydroxylases.
• To provide an efficient and cost-effective solution for hydroxylase engineering.

Main Methods:

• Development of HyS-BST, a dedicated self-trained semisupervised framework tailored for hydroxylases.
• Integration of a self-training strategy with Bayesian dynamic tuning for enhanced prediction accuracy.
• Utilized a limited hydroxylase dataset for training and validation.

Main Results:

• HyS-BST achieved a coefficient of determination (R²) of 0.96 and a Pearson correlation coefficient (PCC) of 0.98 after ten training iterations.
• The model demonstrated a low root mean squared error (RMSE) of 0.06 on the test set.
• Compared to cross-family models, HyS-BST improved PCC and RMSE by approximately 70%, showcasing superior hydroxylase-specific performance.

Conclusions:

• The HyS-BST framework offers a specialized and highly accurate solution for predicting hydroxylase thermostability.
• This approach significantly reduces the search space for enzyme variants and conserves experimental resources.
• HyS-BST represents a cost-effective advancement for hydroxylase engineering and thermostability design.