Local quality assessment in homology models using statistical potentials and support vector machines

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Modeling

Background:

• Accurate local quality assessment is crucial for evaluating protein homology models.
• Existing methods for measuring local structural similarity and quality prediction vary in effectiveness.

Purpose of the Study:

• To identify optimal methods for measuring local structural similarity between homology models and native structures.
• To assess the performance of statistical potentials and machine learning approaches for local model quality prediction.

Main Methods:

• Evaluation of various local geometric similarity measures.
• Comparison of structure-based methods and statistical potentials (e.g., DFIRE).
• Application of support vector machines (SVMs) combining potentials and structural features.

Main Results:

• Two structural superposition methods were identified as best reproducing human expert assessments of local model quality.
• The DFIRE statistical potential performed comparably to the ProQres structure-based method.
• SVMs integrating multiple statistical potentials and structural features showed improved prediction performance.

Conclusions:

• Structural superposition offers a reliable approach for local quality assessment in homology modeling.
• DFIRE and combined machine learning strategies represent advanced methods for predicting local model quality.
• Further development in integrating diverse features can enhance the accuracy of protein model quality evaluation.