Seq2Topt: a sequence-based deep learning predictor of enzyme optimal temperature.

Area of Science:

• Biochemistry
• Computational Biology
• Enzyme Engineering

Background:

• Enzyme optimal temperature (Topt) is crucial for catalytic activity.
• Accurate prediction of Topt is essential for enzyme applications.
• Existing models have limitations in Topt prediction accuracy.

Purpose of the Study:

• Develop a deep learning predictor for enzyme Topt using protein sequences.
• Enhance enzyme mining and in-silico enzyme design.
• Create a versatile prediction platform for enzyme properties.

Main Methods:

• Developed Seq2Topt, a deep learning model utilizing protein sequences.
• Employed multi-head attention to identify key protein regions for Topt.
• Validated Seq2Topt through case studies on thermophilic enzymes and mutation effects.

Main Results:

• Seq2Topt achieved superior accuracy in Topt prediction (RMSE = 12.26°C, R2 = 0.57).
• The model identified critical protein regions influencing Topt.
• Developed accurate predictors for enzyme optimal pH (Seq2pHopt) and melting temperature (Seq2Tm).

Conclusions:

• Seq2Topt is a promising computational tool for enzyme discovery and design.
• The model architecture can be extended to predict other enzyme properties.
• This work lays the foundation for a comprehensive enzyme property prediction platform.