Multi-Omics Graph Attention Network for Key Gene Prediction in Triple-Negative Breast Cancer.

Area of Science:

• Computational biology and bioinformatics
• Genomics and precision medicine
• Cancer research and oncology

Background:

• Triple-negative breast cancer (TNBC) is an aggressive subtype lacking targeted therapies.
• Need for interpretable biomarkers to guide personalized treatment strategies in TNBC.
• Current approaches often analyze single omics data, limiting comprehensive understanding.

Purpose of the Study:

• To develop a multimodal graph attention network (GAT) framework for TNBC.
• To integrate single-cell RNA sequencing (scRNA-seq), single-cell ATAC sequencing (scATAC-seq), and radiomics data.
• To identify robust and interpretable biomarkers for TNBC prognosis and therapeutic vulnerability discovery.

Main Methods:

• Integration of scRNA-seq, scATAC-seq, and radiomics data using a GAT-based multimodal framework.
• Canonical correlation analysis for cross-modal feature alignment.
• Incorporation of intercellular communication and transcription factor binding information into a multimodal graph.
• Ensemble multilayer perceptron with variational dropout for patient prognosis stratification.

Main Results:

• The multimodal GAT framework demonstrated strong predictive performance in an external TCGA-TNBC cohort (log-rank p<0.01).
• Achieved superior prognostic prediction (AUC-ROC = 0.839) compared to single-omics and alternative graph-based methods.
• Validated known TNBC drivers (e.g., PI3K, EGFR) and uncovered novel regulatory pathways (e.g., complement-coagulation, ECM-integrin signaling).

Conclusions:

• The proposed multimodal framework offers an interpretable strategy for biomarker discovery in TNBC.
• Identified both established and novel therapeutic vulnerabilities, advancing precision oncology.
• Provides a scalable approach for integrating multi-omics and imaging data for personalized cancer treatment.