Modeling Enzyme Temperature Stability from Sequence Segment Perspective.

Area of Science:

• Biochemistry
• Computational Biology
• Protein Engineering

Background:

• Enzyme thermal stability is critical for industrial and research applications.
• Experimental methods for determining enzyme thermal properties are resource-intensive.
• Existing computational models face challenges due to limited and imbalanced data.

Purpose of the Study:

• To create a curated dataset for benchmarking enzyme thermal modeling.
• To develop an efficient and accurate deep learning framework for predicting enzyme temperature stability.
• To demonstrate the utility of the model in guiding enzyme engineering efforts.

Main Methods:

• Curated a novel dataset for enzyme temperature stability.
• Developed the Segment Transformer, a deep learning model utilizing segment-level protein sequence representations.
• Evaluated model performance using RMSE, MAE, Pearson, and Spearman correlations.

Main Results:

• The Segment Transformer achieved state-of-the-art performance in predicting enzyme temperature stability.
• The model demonstrated the importance of segment-level features in thermal behavior prediction.
• Engineered a cutinase enzyme with a 1.64-fold improvement in heat-treated activity using 17 mutations.

Conclusions:

• The Segment Transformer offers an efficient and accurate approach to predict enzyme thermal stability.
• The developed dataset and model facilitate advancements in enzyme thermal modeling and engineering.
• The study validates the model's capability to guide protein engineering for enhanced thermostability.