Ranking the quality of protein structure models using sidechain based network properties.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning in Biology

Background:

• Protein structure determination from amino acid sequences is a fundamental challenge in molecular biology.
• Current protein structure prediction methods generate numerous models, necessitating effective ranking strategies.
• Distinguishing native protein structures from inaccurate models is crucial for biological research.

Purpose of the Study:

• To develop and validate a computational method for ranking protein structure models.
• To differentiate between native-like and non-native protein structures using predicted models.
• To provide a tool for selecting high-quality protein structure models.

Main Methods:

• Utilized Protein Structure Networks (PSNs) for both native and modeled proteins.
• Integrated Support Vector Machines (SVMs) with PSNs for quality estimation.
• Employed regression analysis for model ranking and selection.
• Developed a web-server for practical application of the ranking tool.

Main Results:

• The combined PSN and SVM approach effectively estimates protein model quality.
• Regression analysis successfully ranks multiple protein structure models.
• Structures ranked above 16 demonstrate native protein-like characteristics.
• Structures ranked below 10 were identified as non-native.

Conclusions:

• The developed method provides a reliable way to rank protein structure models.
• The tool aids in selecting accurate protein structures from prediction ensembles.
• The web-server offers a practical solution for researchers evaluating protein models.