Prediction of mutation-induced protein stability changes based on the geometric representations learned by a...

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning

Background:

• Protein thermostability is crucial for biological function.
• Predicting mutation-induced stability changes is vital for understanding protein structure-function relationships.
• Accurate prediction aids protein engineering and pharmaceutical design.

Purpose of the Study:

• To present mutDDG-SSM, a deep learning framework for predicting mutation-induced protein stability changes.
• To leverage geometric protein structure representations for enhanced prediction accuracy.
• To provide a valuable tool for protein engineering and drug design.

Main Methods:

• Developed mutDDG-SSM, a two-part deep learning framework.
• Utilized a graph attention network with self-supervised learning for structural feature extraction.
• Employed an eXtreme Gradient Boosting model for stability change prediction to mitigate overfitting.

Main Results:

• mutDDG-SSM demonstrated high performance in predicting mutation effects on protein stability across independent datasets.
• Case studies on myoglobin and p53 validated the model's effectiveness.
• The model showed good unbiasedness, with similar accuracy for direct and inverse mutations.

Conclusions:

• The pre-trained model yields meaningful features for downstream tasks.
• mutDDG-SSM serves as a valuable tool for protein engineering and drug design.
• The framework accurately predicts mutation-induced protein stability changes.