QUATgo: Protein quaternary structural attributes predicted by two-stage machine learning approaches with...

Area of Science:

• Computational biology
• Proteomics
• Structural biology

Background:

• Proteins often function as oligomers, not single chains.
• Predicting protein quaternary structure is crucial for understanding protein function.
• Existing methods struggle with integrating diverse coding and limited data.

Purpose of the Study:

• To develop a novel computational tool, QUATgo, for predicting protein quaternary structures.
• To address limitations in existing methods regarding heterogeneous coding and data scarcity.
• To predict oligomers with more than 12 subunits.

Main Methods:

• Utilized three sequence coding methods: dipeptide composition (novel), protein half-life characteristics, and modified functional domain composition.
• Implemented a two-stage architecture to overcome insufficient single-subunit data.
• Employed 10-fold cross-validation for accuracy assessment.

Main Results:

• QUATgo achieved 49.0% cross-validation accuracy and 31.1% independent test accuracy.
• Demonstrated high accuracy (61.5%) in predicting the quaternary structure of influenza virus hemagglutinin.
• Introduced dipeptide composition as a new feature for quaternary structure prediction.

Conclusions:

• QUATgo offers an effective approach for predicting protein quaternary structures, especially for large oligomers.
• The tool integrates novel coding strategies to improve prediction accuracy with limited data.
• QUATgo is publicly available as a web server for the scientific community.