Effective discrimination between biologically relevant contacts and crystal packing contacts using new determinants

Area of Science:

• Structural biology
• Bioinformatics
• Computational biology

Background:

• X-ray crystallography reveals molecular contacts, categorized as biologically relevant or crystal packing.
• Differentiating these contact types is crucial for accurate structural model interpretation.
• Increasingly large datasets necessitate robust methods for contact classification.

Purpose of the Study:

• To systematically analyze and differentiate biologically relevant contacts from crystal packing contacts.
• To identify distinct properties of biologically relevant contacts.
• To develop a computational model for classifying protein-ligand and protein-protein interactions.

Main Methods:

• Comparative analysis of contact properties, including packing density, amino acid composition, and evolutionary measures.
• Development of a random forest classification model based on identified distinguishing features.
• Validation using 5-fold cross-validation and independent test sets.

Main Results:

• Biologically relevant contacts exhibit tighter packing and more pronounced differences between core and surface regions compared to crystal packing contacts.
• The developed random forest model achieved high performance (ROC 0.923, accuracies 91.4%-91.7%).
• The model outperformed existing methods like DiMoVo, Pita, Pisa, and Eppic.

Conclusions:

• Distinct properties of biologically relevant contacts can be leveraged for accurate classification.
• The developed random forest model offers a reliable tool for distinguishing contact types in structural biology.
• This approach aids in the validation of protein structures, oligomeric states, and protein complexes.