Feature-based multiple models improve classification of mutation-induced stability changes.

Area of Science:

• Computational biology
• Protein engineering
• Bioinformatics

Background:

• Predicting protein stability changes is crucial for computational protein design.
• Current machine learning methods struggle with novel, non-homologous proteins and vary in performance based on mutation site characteristics.
• Mutation site properties like secondary structure and surface accessibility influence prediction accuracy.

Purpose of the Study:

• To develop a more accurate and robust method for predicting protein variant stability changes.
• To address the limitations of existing methods in handling diverse mutation types and unseen proteins.
• To improve the performance of stability change classification across different mutation contexts.

Main Methods:

• Proposed a feature-based multiple model approach named Evolutionary, Amino acid, and Structural Encodings with Multiple Models (EASE-MM).
• Developed five distinct models targeting mutations in exposed, buried, helical, sheet, and coil residues.
• Implemented a consensus classification using two models selected based on predicted accessible surface area and secondary structure.

Main Results:

• EASE-MM achieved a Matthews correlation coefficient of 0.44, a notable improvement over previous work.
• The method correctly classified 73% of stabilizing and 75% of destabilizing protein variants.
• Independent testing on 238 mutations confirmed superior performance compared to related methods.

Conclusions:

• EASE-MM demonstrated superior and more balanced performance across various mutation types compared to existing methods.
• The use of multiple, specialized models with relevant features for specific mutation contexts explains the improved accuracy.
• Results support the hypothesis that distinct interactions influence stability changes in residues based on their environment (exposed/buried) and secondary structure.