A three-state prediction of single point mutations on protein stability changes

Area of Science:

• Biophysics
• Computational Biology
• Protein Science

Background:

• Assessing mutation-induced protein stability changes is fundamental in structural biology.
• Predicting the change in free energy (DeltaDeltaG) upon mutation aids proteomics and genomics annotation.
• Experimental DeltaDeltaG values exhibit uncertainty, with many near zero, complicating mutation-structure relationships.

Purpose of the Study:

• To develop a novel predictor for classifying single point protein mutations.
• To discriminate between stabilizing (DeltaDeltaG > 1.0 kcal/mole), destabilizing (DeltaDeltaG < -1.0 kcal/mol), and neutral (-1.0 <= DeltaDeltaG <= 1.0 kcal/mole) mutations.

Main Methods:

• Utilized a support vector machine (SVM) model.
• Input data derived from either protein sequence or structure information.
• Employed a three-state classification system for mutations.

Main Results:

• Achieved 56% accuracy using sequence information and 61% accuracy with protein structure data.
• Demonstrated a mean value correlation coefficient of 0.27 (sequence) and 0.35 (structure).
• Performance significantly surpassed random prediction by approximately 20 percentage points.

Conclusions:

• The developed method enhances the prediction accuracy of free energy changes from single point mutations.
• Incorporated a hypothesis of thermodynamic reversibility to reframe experimental data.
• Balanced the dataset distribution, mitigating overestimation issues common in methods trained on unbalanced data.