Multi-omics disease module detection with an explainable Greedy Decision Forest.

Area of Science:

• Bioinformatics
• Systems Biology
• Computational Biology

Background:

• Machine learning often overlooks domain knowledge, limiting its application in scientific disciplines like systems biology.
• Integrating network or graph-based domain knowledge can significantly enhance model performance in data-driven research.
• Systems biology leverages machine learning for analyzing multi-modal biomedical data, crucial for advancing precision medicine.

Purpose of the Study:

• To develop versatile, network-based algorithms that incorporate domain knowledge for improved machine learning model performance.
• To introduce an interpretable subnetwork detection method applicable across various scientific domains.
• To demonstrate the utility of explainable AI in uncovering disease mechanisms from multi-omics data.

Main Methods:

• Developed a novel Greedy Decision Forest (GDF) algorithm for subnetwork detection.
• Utilized multi-modal node features within a network context.
• Focused on inherent interpretability of the GDF model to foster expert trust and adoption.

Main Results:

• Successfully demonstrated subnetwork detection using the novel GDF approach.
• The inherent interpretability of the GDF facilitates expert understanding and trust.
• The methodology is applicable to diverse research areas, with a focus on biomedicine.

Conclusions:

• The proposed explainable Greedy Decision Forest offers a powerful tool for integrating domain knowledge into machine learning.
• This approach can uncover disease-causing network modules from multi-omics data, advancing the understanding of complex diseases.
• The method supports the goals of precision medicine by enabling system-level modeling of patient complexity.