QUATgo

QUATgo: Machine Learning Prediction of Protein Quaternary Structure

QUATgo predicts protein quaternary structural attributes and oligomeric states, including complexes with over 12 subunits, from amino acid sequences using heterogeneous feature encoding and a two-stage machine learning architecture.

Key Features:

• Heterogeneous Feature Encoding: Integrates dipeptide composition, protein half-life characteristics, and modified functional domain composition coding to represent amino acid interactions, protein stability, and domain architecture while reducing feature vector dimensionality.
• Two-Stage Machine Learning Architecture: Improves prediction robustness for limited or incomplete sequence data, including single-chain inputs from protein oligomers.
• Model Evaluation: Uses 10-fold cross-validation with 49.0% cross-validation accuracy and 31.1% independent test accuracy.
• Case Study Validation: Achieves 61.5% accuracy in predicting the quaternary structure of influenza virus hemagglutinin proteins.

Scientific Applications:

• Protein Structure Prediction: Predicts oligomeric states and quaternary structural organization for structural biology and proteomics studies.
• Viral Protein Analysis: Supports quaternary structure prediction of viral proteins, including influenza virus hemagglutinin.

Methodology:

Encodes protein sequences using dipeptide composition, protein half-life characteristics, and modified functional domain composition coding, followed by classification through a two-stage machine learning framework. Model performance is assessed using 10-fold cross-validation and independent testing datasets.