GAIN-BRCA: a graph-based AI-net framework for breast cancer subtype classification using multiomics data

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Accurate breast cancer subtyping is crucial for prognosis and treatment.
• Existing machine learning models often fail to leverage multiomic data effectively.
• Graph-based integration of omics data remains underexplored for capturing biological associations.

Purpose of the Study:

• To develop a novel graph-based method (GAIN-BRCA) for integrating multiomic datasets (mRNA, DNA methylation, miRNA) from breast cancer patients.
• To improve the accuracy of breast cancer subtype prediction by capturing biological context through feature interactions.
• To identify subtype-specific prognostic biomarkers for precision therapeutics.

Main Methods:

• Developed GAIN-BRCA, a graph-based machine learning framework.
• Integrated native features from mRNA, DNA methylation (CpG), and miRNA data.
• Synthesized features from miRNA-mRNA and CpG-mRNA interactions to compute weights, creating a transformed feature vector.

Main Results:

• GAIN-BRCA achieved superior performance with an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.98 and an accuracy of 0.92.
• Outperformed existing methods MOGONET (0.72 accuracy) and moBRCA-net (0.86 accuracy).
• Identified subtype-specific prognostic genes (e.g., KRAS, TOX, MITF, TOB1) and biomarkers using GAIN-BRCA and SHAP analysis.

Conclusions:

• GAIN-BRCA effectively integrates multiomic data for accurate breast cancer subtyping and prognosis.
• The method identifies novel subtype-specific biomarkers, paving the way for precision medicine.
• The GAIN-BRCA code is publicly available for further research and application.