Hierarchical Multi-Label Classification With Gene-Environment Interactions in Disease Modeling

Area of Science:

• Biomedical Informatics
• Genetics
• Computational Biology

Background:

• Gene-environment (G-E) interactions are crucial for understanding disease etiology beyond individual genetic or environmental factors.
• Existing G-E interaction analyses often overlook hierarchical multi-label classification and semi-supervised learning scenarios.
• Unlabeled data, common in real-world biomedical studies, are frequently excluded by current hierarchical multi-label classification methods.

Purpose of the Study:

• To develop a novel computational framework for analyzing G-E interactions within a hierarchical multi-label classification context.
• To address the challenge of incorporating unlabeled data in G-E interaction analysis using a semi-supervised approach.
• To provide a robust method for identifying complex disease outcomes influenced by G-E interactions.

Main Methods:

• A novel approach for two-layer hierarchical response incorporating G-E interactions was developed.
• A two-step penalized estimation strategy was employed.
• An efficient expectation-maximization (EM) algorithm was utilized for parameter estimation.

Main Results:

• The proposed method demonstrated superior performance in both classification accuracy and feature selection compared to existing approaches.
• Simulations confirmed the effectiveness and robustness of the developed algorithm.
• Application to The Cancer Genome Atlas (TCGA) lung cancer data validated its practical utility.

Conclusions:

• This study bridges a significant knowledge gap in G-E interaction analysis by introducing a versatile framework for hierarchical multi-label classification.
• The developed method offers a powerful tool for dissecting complex disease mechanisms influenced by G-E interactions.
• The approach effectively handles semi-supervised learning scenarios, enhancing applicability to real-world biomedical data.