PROTS-RF: a robust model for predicting mutation-induced protein stability changes.

Area of Science:

• Protein Engineering
• Computational Biology
• Biophysics

Background:

• Improving protein thermostability is crucial for scientific research and practical applications.
• Predicting thermostability changes from mutations is a key challenge in protein engineering.

Purpose of the Study:

• To develop and validate PROTS-RF, a robust Random Forest model for predicting protein thermostability changes.
• To assess the model's capability in predicting effects of single, double, and multiple point mutations.

Main Methods:

• Utilized a Random Forest algorithm.
• Incorporated 41 features: evolutionary information, secondary structure, solvent accessibility, and fragment-based features.
• Evaluated model performance using accuracy and ROC curves on single, double, and multiple point mutation datasets.

Main Results:

• PROTS-RF achieved high prediction accuracies (0.799 for single, 0.782 for double, 0.787 for multiple mutations).
• Achieved excellent ROC AUC scores (0.873, 0.868, 0.862 respectively).
• Demonstrated that a model trained on single mutations can effectively predict multiple mutations and showed robustness with reverse mutations.

Conclusions:

• PROTS-RF is a robust and accurate model for predicting protein thermostability changes.
• The model's ability to handle multiple mutations challenges previous assumptions.
• Testing datasets based on physical principles enhance the evaluation of predictive model robustness.