DDMut: predicting effects of mutations on protein stability using deep learning.

Area of Science:

• Computational Biology
• Protein Engineering
• Bioinformatics

Background:

• Predicting mutation effects on protein stability is vital for biotechnology and medicine.
• Existing tools face limitations in speed, accuracy, and bias towards destabilizing mutations.

Purpose of the Study:

• To develop a fast and accurate deep learning model for predicting changes in Gibbs Free Energy (ΔΔG) upon single and multiple point mutations.
• To improve upon existing methods by addressing limitations in computational time, predictive power, and prediction bias.

Main Methods:

• Developed DDMut, a siamese network integrating graph-based 3D environment representations with convolutional layers and transformer encoders.
• Utilized both forward and hypothetical reverse mutations to ensure model anti-symmetry.
• Trained deep learning models on extensive datasets of protein mutations and their stability changes.

Main Results:

• DDMut achieved high accuracy, with Pearson's correlations of up to 0.70 for both single (RMSE: 1.37 kcal/mol) and multiple point mutations (RMSE: 1.84 kcal/mol).
• The model demonstrated superior performance compared to existing methods on non-redundant blind test sets.
• DDMut showed excellent scalability and anti-symmetric prediction for both stabilizing and destabilizing mutations.

Conclusions:

• DDMut provides a computationally efficient and accurate method for predicting mutation-induced changes in protein stability.
• This tool can significantly aid in variant interpretation, protein engineering, and understanding mutation-driven functional consequences.
• DDMut is accessible as a free web server and API for broader scientific use.