MODA: a graph convolutional network-based multi-omics integration framework for unraveling hub molecules and disease...

Area of Science:

• Systems biology
• Bioinformatics
• Computational biology

Background:

• Integrating multi-omics data is crucial for systems biology but challenging due to data complexity and heterogeneity.
• Existing methods often struggle with prior knowledge integration and noise reduction in omics datasets.

Purpose of the Study:

• To develop a robust framework, Multi-Omics Data Integration Analysis (MODA), for effective multi-omics data integration.
• To identify hub molecules, critical pathways, and elucidate biological mechanisms underlying diseases.
• To enhance biological interpretability and stability in omics data analysis.

Main Methods:

• MODA framework leverages multiple machine learning approaches to transform raw omics data into a feature importance matrix.
• Integrates prior knowledge via a biological knowledge graph to mitigate data noise.
• Employs graph convolutional networks with attention mechanisms and overlapping community detection for module extraction.

Main Results:

• MODA outperforms seven existing methods in classification performance and demonstrates superior stability across pan-cancer datasets.
• Identified carnitine and palmitoylcarnitine, regulated by BBOX1, as key players in prostate cancer progression.
• Validation through population samples and in vitro experiments confirmed the findings.

Conclusions:

• MODA is an efficient and cost-effective tool for uncovering novel disease mechanisms from multi-omics data.
• The framework advances precision medicine by providing deep biological insights and identifying potential therapeutic targets.
• Highlights the potential of integrating prior knowledge and advanced machine learning for complex biological data analysis.