Use B-factor related features for accurate classification between protein binding interfaces and crystal packing...

Area of Science:

• Structural bioinformatics
• Computational biology
• Biophysics

Background:

• Accurate classification of protein structures is crucial for structural bioinformatics.
• Distinguishing true protein interactions from crystal packing contacts is challenging due to unannotated and false annotations.
• Existing tools face performance limitations with diverse interface sizes.

Purpose of the Study:

• To develop novel features for improved classification of biological interfaces versus crystal packing contacts.
• To investigate the utility of B factor-related features in this classification task.

Main Methods:

• Proposed three B factor-related features: sum of normalized B factors, average interfacial B factor per residue, and average interfacial atoms with negative normalized B factor per residue.
• Investigated feature distributions on multiple datasets, including those with known binding affinity.
• Compared performance against interface area and existing prediction methods.

Main Results:

• B factor-related features significantly outperform interface area in distinguishing biological from crystal packing interfaces.
• Demonstrated remarkable cross-dataset classification performance across various datasets.
• The proposed features showed superior results compared to existing prediction methods.

Conclusions:

• B factor-related features are more effective than interface area for classifying protein-protein interfaces.
• The developed computational methods show potential for accurate, large-scale identification of biological interactions from PDB data.
• These findings can enhance the reliability of structural bioinformatics studies.