Bayesian Network Models for Genetics and Metabolomics Studies of Fetal Programming

项目总结

Project Summary Integrated modeling of genetics and metabolomics data demands careful specification of complex dependencies, particularly when investigating mechanisms underlying fetal programming of related clinical phenotypes in mothers and their newborns. Recent efforts to identify `genetically determined metabotypes' using classic genome wide association study approaches have provided initial links between genetics and metabolomics data; however, to accurately characterize genetic and metabolic contributions to fetal development, more sophisticated models are required. The goal of this study is to develop Bayesian network models for cogent synthesis of genetics and metabolomics data related to clinical phenotypes for mothers and their newborns. In networks, nodes represent omics features of interest and edges represent relationships among them. Bayesian networks construct a series of directed relationships among nodes in which a variable represented by a `child' node is described conditional on its `parent' nodes. It is our hypothesis that metabotyping in mothers and newborns will provide a set of candidate gene-metabolite relationships whose joint contribution to clinical phenotype(s) can be parsimoniously modeled in Bayesian networks that also incorporate dependencies among genotypes and metabolites. To develop network models, we will use existing genetics and metabolomics data for mother/newborn pairs from the Hyperglycemia and Adverse Pregnancy Outcome (HAPO) Study. HAPO was an international, observational study involving >23,000 pregnant mothers and their newborns from 2000-2006 that demonstrated a positive association between maternal glucose levels during pregnancy and newborn birth weight and adiposity. HAPO Metabolomics is an ongoing NIH-funded investigation of mother/newborn pairs of Northern European, Afro-Caribbean, Mexican-American and Thai ancestry involving targeted and non-targeted gas-chromatography/mass-spectrometry profiling of maternal and newborn cord serum. HAPO Genetics studies involved genome-wide genotyping for HAPO mother/newborn pairs in the same four ancestry groups. ~1400 mother/child pairs are represented in both HAPO Metabolomics and HAPO Genetics studies. Using these data, we will apply a comprehensive metabotyping pipeline to identify candidate gene-metabolite relationships for integrated models. We will then develop and apply Bayesian network models to unify maternal and newborn metabotypes in conjunction with complex dependencies among maternal and newborn genotypes and metabolic profiles. These proposed analyses will augment existing metabotyping approaches by integrating metabotypes into Bayesian network models for fuller characterization of maternal and fetal genetic and metabolic underpinnings of known clinical phenotype associations. The proposed methods will also be more broadly applicable to integration of diverse omics data for related individuals.