XL₁R-Net: Explainable AI-driven improved L₁-regularized deep neural architecture for NSCLC biomarker identification

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Non-small cell lung cancer (NSCLC) is molecularly diverse, necessitating biomarker identification for subtyping and targeted therapy.
• Copy number variation (CNV) data, reflecting genomic instability, is underexplored for NSCLC biomarker discovery.

Purpose of the Study:

• To identify novel NSCLC biomarkers using copy number variation (CNV) data.
• To leverage explainable AI (XAI) for biomarker discovery and classification.

Main Methods:

• Developed an eXplainable AI (XAI)-driven L1-regularized deep learning architecture (XL1R-Net) for NSCLC subtyping.
• Utilized XAI for feature identification to uncover NSCLC-relevant biomarkers from CNV data.

Main Results:

• Achieved 84.95% classification accuracy using a Multilayer Perceptron (MLP) model with 10-fold cross-validation.
• Identified twenty NSCLC-relevant biomarkers, with twelve deemed potentially druggable and eighteen associated with patient survival.
• Validated nine biomarkers with existing literature and five with the OncoKB Gene List.

Conclusions:

• Identified seven novel biomarkers warranting investigation for NSCLC therapy.
• Highlights the need for multiomics data integration to fully capture NSCLC heterogeneity.
• Suggests future research directions focusing on multiomics data integration for better understanding of NSCLC.