Prediction of enzyme mutant activity using computational mutagenesis and incremental transduction

Area of Science:

• Computational biology
• Enzyme engineering
• Machine learning

Background:

• Wet laboratory mutagenesis for enzyme activity prediction is costly and time-consuming.
• Existing computational methods often lack adaptability to accumulating experimental data.
• There is a need for efficient and accurate prediction models in enzyme mutagenesis.

Purpose of the Study:

• To propose an incremental transductive learning method (T2bRF) for predicting enzyme mutant activity.
• To enhance the efficiency and accuracy of enzyme mutagenesis predictions.
• To integrate Delaunay tessellation and 4-body statistical potentials within a machine learning framework.

Main Methods:

• Developed an incremental transductive random forest (T2bRF) model.
• Utilized Delaunay tessellation for data representation.
• Incorporated 4-body statistical potentials for feature extraction.
• Employed cross-validation for experimental validation.

Main Results:

• T2bRF achieved 90% accuracy on T4 and LAC datasets, and 86% on HIV-1.
• The incremental transductive approach significantly outperformed standard one-shot learning methods.
• Results demonstrate superior performance compared to state-of-the-art methods yielding 80% or less.

Conclusions:

• The proposed T2bRF method offers a more efficient and accurate approach to predicting enzyme mutant activity.
• Incremental learning effectively handles cumulative experimental data in enzyme engineering.
• T2bRF represents a significant advancement in computational enzyme mutagenesis prediction.