Explainable artificial intelligence for multi-modal cancer analysis: From genomics to immunology

Area of Science:

• Oncology
• Bioinformatics
• Artificial Intelligence

Background:

• Cancer is a complex, heterogeneous disease with challenges in prognosis and treatment prediction.
• Single-modality approaches fail to capture the full biological landscape of cancer.
• Multifactorial complexity arises from tumor-microenvironment interactions, immune modulation, and therapeutic pressures.

Purpose of the Study:

• To review the application of selective multimodal deep learning (MDL) in precision oncology.
• To highlight the integration of diverse biomedical data for improved cancer analysis.
• To emphasize the role of explainable AI (XAI) in understanding complex cancer models.

Main Methods:

• Integration of complementary biomedical data (genomics, transcriptomics, imaging, EHRs, etc.) using MDL.
• Mechanism-informed fusion strategies to capture cross-scale dependencies and emergent patterns.
• Application of explainable AI (XAI) for transparent, biologically grounded model interpretation.

Main Results:

• MDL models can capture intricate cross-scale dependencies and emergent patterns in cancer.
• Immunology-informed integration enhances biomarker discovery and immunotherapy stratification.
• XAI provides biologically grounded explanations for complex computational predictions.

Conclusions:

• Selective MDL offers a transformative framework for personalized precision oncology.
• Rigorous validation, including statistical metrics and biological plausibility, is crucial.
• Future directions include federated learning, causal inference, and digital twins for advancing personalized cancer care.