Protein design using structure-based residue preferences.

Area of Science:

• Protein design
• Computational biology
• Biophysics

Background:

• Modern protein design utilizes large neural networks, but critical residue dependencies remain unclear.
• Understanding these dependencies is key to predicting protein function and guiding design efforts.

Purpose of the Study:

• To determine if individual amino acid preferences, independent of mutation interactions, can predict combinatorial mutation effects.
• To develop a computationally efficient method for predicting mutation effects based on local structural context.

Main Methods:

• Analysis of 8 diverse datasets to quantify the predictive power of single-residue preferences.
• Development of CoVES (Combinatorial Variant Effects from Structure), an unsupervised method leveraging local structural contexts.
• Comparison of CoVES performance against model-free and complex computational methods.

Main Results:

• Single amino acid preferences explained a significant portion (R² ~ 78-98%) of combinatorial mutation effects across datasets.
• Accurate prediction of held-out variant effects was achieved with limited data (Pearson r > 0.80).
• CoVES demonstrated superior performance compared to model-free approaches and comparable results to complex models.

Conclusions:

• Individual residue preferences are powerful predictors of protein function, simplifying complex mutation effect predictions.
• CoVES provides an effective and computationally accessible alternative for identifying functional protein mutations.
• The findings offer a more streamlined approach to protein design and engineering.