An explainable AI-driven biomarker discovery framework for Non-Small Cell Lung Cancer classification

Area of Science:

• Biomedical Informatics
• Computational Biology
• Artificial Intelligence in Oncology

Background:

• Non-Small Cell Lung Cancer (NSCLC) presents molecular heterogeneity, necessitating precise subtype classification for effective treatment.
• Lung Adenocarcinoma (LUAD) and Lung Squamous Cell Carcinoma (LUSC) are the primary NSCLC subtypes with distinct molecular profiles.
• Accurate subtyping of NSCLC is crucial for developing targeted therapies and improving patient outcomes.

Purpose of the Study:

• To develop a novel explainable AI (XAI)-based deep learning framework for discovering key biomarkers in NSCLC.
• To identify a minimal yet effective set of biomarkers for differentiating between LUAD and LUSC.
• To explore the potential of discovered biomarkers in targeted therapy and patient survivability prediction.

Main Methods:

• An autoencoder was used to reduce feature dimensionality, followed by a feed-forward neural network for NSCLC classification.
• An XAI module was integrated to identify significant biomarkers from the deep learning model.
• Machine learning models were trained and validated using the discovered biomarkers for NSCLC subtyping.

Main Results:

• The XAI framework identified 52 relevant biomarkers for NSCLC subtype classification.
• A Multilayer Perceptron model achieved 95.74% accuracy in classifying NSCLC subtypes using these biomarkers.
• 14 of the 52 biomarkers were identified as druggable, and 28 were associated with patient survivability prediction.

Conclusions:

• The proposed XAI framework effectively identifies NSCLC biomarkers with high classification accuracy.
• The discovered biomarkers hold potential for targeted therapy development and predicting patient outcomes.
• Seven novel biomarkers were identified for NSCLC subtyping, warranting further investigation for their therapeutic implications.