Gene-Embedded Multi-Modal Networks for Population-Scale Multi-Omics Discovery.

Area of Science:

• Systems Biology
• Metabolomics
• Transcriptomics

Background:

• Small molecules (SMs) are crucial for biological processes but their downstream effects on pathways and gene expression are poorly understood.
• Integrating untargeted metabolomics and sequencing data is challenging due to scale, variability, and noise differences.

Purpose of the Study:

• To develop a multi-omics framework for harmonizing and analyzing diverse biological data.
• To construct robust transcriptome-SM interaction networks (TSI-Nets) for discovering SM functions and complex trait associations.

Main Methods:

• Developed a machine learning protocol for data harmonization (batch correction, PCA, regression).
• Proposed a semi-supervised network inference method using meta-analysis to build TSI-Nets.
• Applied the framework to analyze insulin sensitivity (IS) and identified key molecular players.

Main Results:

• The framework successfully integrated metabolomic, lipidomic, and transcriptomic data, outperforming conventional methods in network generation.
• Network analysis revealed novel connections between SMs and genes related to IS, surpassing single-omics findings.
• Identified a pathway linking N-acetylglycine (NAG) from gut microbiota to immune regulation, improving IS and reducing obesity/inflammation.

Conclusions:

• The developed framework provides a scalable solution for multi-modal network inference and SM pathway discovery.
• TSI-Nets offer valuable insights into the implications of SMs for human health and complex traits.
• The generated networks are publicly available, serving as a resource for the research community.