Network Models for Metabolomics

Summary
Our proposal describes network based approaches for the analysis of data from metabolomics studies. The specific
aims of this proposal include:
Aim 1: Variable selection methods in metabolomics studies, incorporating metabolite dependence and external
pathway information. We propose a Bayesian variable selection approach to incorporate both a partially observed,
external pathway network and a data-driven partial correlation network.
Aim 2: Models to identify differential metabolic networks that characterize groups within a study, and addi-
tionally detect subcomponents with group-specific associations with an outcome. When metabolic networks
differ according to groups, exposure levels (e.g. treatment) or other factors, our proposed framework will provide an
approach to identify group-specific networks as well as subcomponents that are associated with outcome, in a possibly
group-specific manner.
Aim 3: Methods to identify metabolite subnetworks that collectively mediate the relationship between an ex-
posure and an outcome. We propose a two-phase analysis framework involving (1) Detection of metabolite subnet-
works enriched for association with the outcome; and (2) Estimation of the magnitude of the indirect effects mediated
by metabolite subnetworks.
Application to testing clinical hypotheses in the WHI, NHS and HAPO metabolomics studies: Using methods
developed in Aims 1, we will identify metabolites and modules associated with risk of stroke in the NHS and maternal
metabolomic markers of newborn adiposity in the HAPO study. Using methods in Aim 2, in the WHI, we will identify
metabolic subnetworks that change due to initiation of hormone therapy (estrogen, progestin plus estrogen, placebo)
within age groups, with treatment/age dependent modules associated with subsequent risk of CHD; in the HAPO
study, detect maternal metabolite networks that differ between mothers of boys versus mothers of girls and sex-specific
subcomponents that inform sex-related differences in newborn body composition related to maternal glycemia during
pregnancy. Aim 3 methods will be applied to detect metabolite subnetworks that potentially mediate the association of
exposures such as dietary score and risk of CHD in the WHI; and maternal glucose during pregnancy and newborn
adiposity in HAPO.
IMPACT: Significant federal investment has been made into research of the metabolomic underpinnings of complex
disorders, such as through the NIH's Common Fund Metabolomics program. Our interdisciplinary team proposes to
develop and apply new statistical models to effectively mine rapidly growing metabolomics data sources to elucidate the
etiology of complex disorders such as CHD, stroke and maternal glycemia during pregnancy as it relates to newborn
size at birth.

Grant Number: 5R01LM013444-04
NIH Institute/Center: NIH
Principal Investigator: RAJI BALASUBRAMANIAN