Metabolic Network SGL

Metabolic Network SGL integrates metabolic network features with phenotypic data using sparse group lasso (SGL) to identify extreme currents (ECs) and sub-pathways associated with clinical outcomes.

Key Features:

• Integration with '-omics' data and KEGG: Combines metabolic network data from the KEGG database with gene expression and other '-omics' datasets to link network features to phenotypes.
• Decomposition into Extreme Currents (ECs): Decomposes metabolic networks into ECs under steady-state conditions with non-negative flux constraints to define sub-pathways.
• Sparse Group Lasso (SGL): Implements SGL to enforce group-level selection and within-group sparsity, enabling selection of sparse sets of feature groups and sparse features within groups.
• Feature clustering and correlation-based grouping: Defines features as clusters of ECs and organizes feature groups based on correlations among these clustered features.
• Application to clinical outcomes: Applied to associate metabolic network features with tumor versus normal states in prostate cancer and with survival time in glioblastoma multiforme.
• Performance versus existing methods: Simulations reported superior performance compared with methods such as the global test for identifying phenotype-associated ECs.

Scientific Applications:

• Oncology research: Enables identification of metabolic sub-pathways associated with cancer phenotypes, exemplified by analyses of prostate cancer and glioblastoma multiforme.
• Gene set analysis: Provides a self-contained approach to gene set analysis by mapping gene expression to EC-based pathway features for phenotype association testing.

Methodology:

Decompose metabolic networks into extreme currents under steady-state conditions with non-negative flux constraints. Apply sparse group lasso to gene expression–derived features to identify phenotype-associated ECs. Use feature clustering and correlation-based grouping to define feature groups for SGL.