Improving Enzyme Optimum Temperature Prediction with Resampling Strategies and Ensemble Learning.

Area of Science:

• Biotechnology
• Enzyme kinetics
• Machine learning

Background:

• Accurate prediction of optimal catalytic temperature (Topt) is vital for enzyme applications in biotechnology.
• Existing machine learning methods like TOME are limited by training data distribution, showing higher errors for thermostable enzymes.
• The TOME model was trained on data with a median Topt of 37 °C, with less than 5% of values above 85 °C.

Purpose of the Study:

• To improve the prediction accuracy of the TOME machine learning method for enzyme optimal temperatures, particularly for thermostable enzymes.
• To address the data imbalance issue in the training dataset that limits TOME's performance on high Topt values.
• To develop and release an improved, publicly available method for predicting enzyme Topt.

Main Methods:

• Application of ensemble learning strategies to enhance predictive modeling.
• Implementation of resampling techniques to address data imbalance in the training set.
• Development of a revised machine learning model named TOMER (temperature optima for enzymes with resampling).

Main Results:

• A 60% decrease in prediction error for high Topt values (>85 °C).
• An overall increase in the model's R-squared value from 0.527 to 0.632.
• Significant improvement in predictive accuracy for thermostable enzymes.

Conclusions:

• The revised TOMER method demonstrates enhanced predictive capabilities for enzyme optimal temperatures, especially for thermostable enzymes.
• Ensemble learning and resampling strategies effectively mitigate data imbalance issues in machine learning models for enzyme Topt prediction.
• The TOMER method and associated resampling strategies are available as open-source Python packages on GitHub for broader research use.