Assessing computational methods for predicting protein stability upon mutation: good on average but not in the...

Area of Science:

• Computational biology
• Protein engineering
• Biophysics

Background:

• Protein modeling and design rely on accurate energy functions to calculate free energy.
• Estimating mutation effects on protein stability and binding affinity is a key challenge.

Purpose of the Study:

• To evaluate the performance of six computational methods in predicting changes in protein stability (DeltaDeltaG) upon single-site mutation.
• To assess the accuracy and precision of these methods against experimental data.

Main Methods:

• Six protein stability prediction methods (CC/PBSA, EGAD, FoldX, I-Mutant2.0, Rosetta, Hunter) were tested.
• Performance was evaluated on a dataset of 2156 mutations, excluding training data.
• Correlation coefficients between experimental and predicted DeltaDeltaG values were calculated.

Main Results:

• Prediction accuracy varied, with correlation coefficients ranging from 0.26 to 0.59.
• All methods demonstrated a correct trend but failed to provide precise DeltaDeltaG values.
• Experimental data showed high reproducibility (correlation coefficient of 0.86).

Conclusions:

• Current computational methods offer valuable trends but require significant improvement for precise stability change predictions.
• Combining methods did not substantially enhance prediction accuracy.
• Advancements in protein force fields are essential for improving computational protein design and stability prediction.