From Black Box to Biological Insight: AttentioFuse Unlocks Multi-Omics Dynamics in Lung Cancer.

Area of Science:

• Computational biology
• Genomics
• Machine learning in oncology

Background:

• Non-small cell lung cancer (NSCLC) subtypes, lung adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC), have distinct molecular profiles impacting prognosis and treatment.
• Deep learning models offer high predictive accuracy for NSCLC but lack transparency, hindering clinical adoption.

Purpose of the Study:

• To develop AttentioFuse, an interpretable deep learning framework for precise prognosis and therapy in LUAD and LUSC.
• To integrate multi-omics data using a Reactome-guided strategy for enhanced biological insight.

Main Methods:

• AttentioFuse utilizes dual-phase learning with omics-specific encoders and hierarchical attention mechanisms for dynamic layer contribution analysis.
• Integrated explainability combines DeepSHAP and global attention weights for gene-to-pathway interpretation.
• Two variants, AttentioFuse-3F and AttentioFuse-5X, were evaluated for discrimination and hierarchical interpretation.

Main Results:

• AttentioFuse achieved state-of-the-art TNM staging performance on TCGA LUAD/LUSC cohorts.
• Identified key biological insights including pan-NSCLC AKT/mTOR metabolic regulation and histology-specific Notch signaling.
• Uncovered pathways related to developmental reactivation, microbiota-associated metastasis, and extracellular matrix remodeling.

Conclusions:

• AttentioFuse-5X successfully balances predictive accuracy with hierarchical, pathway-level explanations.
• The framework transforms black-box deep learning predictions into actionable, biologically informed decision support for oncology.