Generalizable and scalable protein stability prediction with rewired protein generative models

Area of Science:

• Biochemistry
• Computational Biology
• Machine Learning

Background:

• Predicting protein thermostability changes from amino acid substitutions is crucial for disease research and protein engineering.
• Existing protein generative models show promise but their potential for stability prediction is underexploited.

Purpose of the Study:

• To present SPURS, a novel deep learning framework for accurate and efficient protein stability prediction.
• To leverage and integrate complementary protein generative models for enhanced performance.
• To enable broad applications in protein informatics and disease research.

Main Methods:

• SPURS integrates a protein language model and an inverse folding model.
• The unified framework is fine-tuned using large-scale thermostability data.
• Deep learning techniques are employed for stability prediction and analysis.

Main Results:

• SPURS achieves accurate, efficient, and scalable predictions of protein stability.
• The framework generalizes well to novel proteins and mutations.
• SPURS demonstrates utility in zero-shot functional residue identification and protein fitness prediction.

Conclusions:

• SPURS establishes a versatile tool for advancing protein stability prediction and engineering.
• The framework facilitates systematic analysis of stability-pathogenicity links in human diseases.
• SPURS enhances capabilities in protein informatics and practical protein design.