Recon8D: A metabolic regulome network from oct-omics and machine learning.

Area of Science:

• Biochemistry
• Genomics
• Systems Biology

Background:

• Metabolic regulatory networks are complex and incomplete, hindering metabolome prediction from other omics data.
• Understanding these networks is crucial for cancer research and therapeutic development.

Purpose of the Study:

• To predict metabolomic variation in cancer cell lines using machine learning and multi-omics data.
• To identify key molecular features and pathways that drive metabolic changes in cancer.

Main Methods:

• Utilized machine learning to integrate genomics, epigenomics (histone PTMs, DNA methylation), transcriptomics, RNA splicing, miRNA-omics, proteomics, and phosphoproteomics data from ~1000 cancer cell lines.
• Reconstructed multi-omic interaction subnetworks for predictable metabolites.

Main Results:

• The metabolome is strongly associated with the transcriptome; miRNAs, phosphoproteins, and histone PTMs provide the most metabolic information per feature.
• Peripheral metabolites are predictable by enzyme levels, while central metabolites require combinatorial predictors from signaling and redox pathways.
• YAP1 signaling was identified as a top global predictor across four omic layers.

Conclusions:

• Multi-omics data can effectively predict metabolomic variation in cancer cell lines.
• Identified key molecular predictors and subnetworks, highlighting YAP1 signaling as a significant factor.
• Prioritized predictive features for advanced metabolomics assays and identified potential therapeutic targets including synthetic-lethal interactions.