Bayesian Network Biomarker Selection in Metabolomics Data

代谢组学是高通量生物学的主要领域之一。液体代谢组图谱分析 色质联用(LC/MS)可同时测定数千种代谢物。这个 LC/MS代谢组谱数据由于几个特点而提出了独特的挑战，包括固有的 将特征与已知代谢物匹配时的不确定性，真零和缺失值的混合，以及 截然不同的数据分布和依赖模式，阻碍了与其他类型的高 维度数据。在这项研究中，我们计划通过开发贝叶斯分层模型来解决这些问题 融合匹配不确定性的网络标记选择，一种综合的回归框架 对多方组学网络的分析，以及一种新的建模策略，以应对 代谢网络中的缺失值。我们将把新开发的方法应用于大规模、高影响的 代谢组学和转录组数据，以获得新的生物学见解，并提供易于使用的软件 社区。

Metabolomics is one of the major areas of high-throughput biology. Metabolomic profiling by liquid chromatography-mass spectrometry (LC/MS) measures thousands of metabolites at the same time. The LC/MS metabolomic profiling data poses unique challenges due to several characteristics including the intrinsic uncertainty in matching features to known metabolites, the mixing of true zeroes and missing values, and distinct data distribution and dependency patterns that hamper integrative analysis with other types of high- dimensional data. In this study, we plan to tackle the problems by developing Bayesian hierarchical models for network marker selection that incorporates matching uncertainties, a regression framework for integrative analysis of multipartite omics networks, and a novel modeling strategy to address the unique challenge of missing values in the metabolic network. We will apply newly developed methods to large-scale, high-impact metabolomics and transcriptomics data to derive new biological insights, and provide easy-to-use software for the community.