A unified predictor of protein stability changes across all mutation types via implicit structure learning

Area of Science:

• Computational Biology
• Protein Engineering
• Bioinformatics

Background:

• Accurate prediction of protein stability changes is vital for protein engineering.
• Existing models often fail with multi-point mutations and indels due to simplified assumptions.
• Current methods struggle to model backbone conformational changes associated with mutations.

Purpose of the Study:

• To develop an advanced framework, UniStab, for predicting protein stability changes across all mutation types.
• To overcome limitations of current models in handling complex mutations and indels.
• To provide a tool that facilitates rational protein engineering and design of stabilized variants.

Main Methods:

• UniStab utilizes a pre-trained protein folding model for implicit geometric reasoning.
• The framework avoids explicit structure generation, reducing computational cost.
• It captures non-additive epistatic interactions and local backbone rearrangements.

Main Results:

• UniStab achieves state-of-the-art performance on a comprehensive benchmark dataset.
• The model shows particular strength in predicting stability changes for multi-point mutations and indels.
• UniStab offers interpretable structural insights beyond mere predictive accuracy.

Conclusions:

• UniStab represents a significant advancement in predicting protein stability changes for diverse mutation types.
• The framework's ability to model complex interactions and backbone changes enhances its utility in protein engineering.
• UniStab effectively guides the design of stabilized protein variants, advancing rational protein design.