Metabolite levels in human blood are regulated by a relatively strict system of homeostatic control. Previous investigations of homeostasis have taken a number of approaches, and models of glucose and a few other metabolites have been developed, typically focused on a single organ. However, while potentially extremely useful, an accurate and quantitative model of blood metabolite levels under homeostasis does not currently exist. It is well known that numerous demographic and clinical factors such as gender, age, BMI, smoking, etc., as well as pre-analytical factors and many diseases, significantly affect the levels of blood metabolites. Numerous studies in the field of metabolomics have attempted to account for the effects of many such factors. However, efforts to quantify these effects and validate them across different studies have so far been challenging, and resulted in consistent failures to validate discovered putative biomarkers. The challenges to integrate metabolite profiles with clinical and demographic factors are complicated by the high dimensionality of the data and the numerous correlations among the metabolites. Traditional statistical methods are incapable of accounting for these factors, and hence, investigations suffer from a high false discovery rate (FDR). To overcome these challenges, we propose to develop quantitative statistical models of blood metabolite levels in healthy adults, and thereby produce a predictive model of homeostasis. Our preliminary work indicates that we can predict metabolite levels with much reduced variance using the reproducibly measured levels of a large pool of blood metabolites and clinical and demographic variables. We propose to develop sophisticated models of homeostasis based on advanced statistical methods and evaluate their predictive performance across different sample sets and metabolite classes. The proposed project has four main Aims: (1) Obtain broad-based metabolomics data on blood samples collected from geographically distinct sites to explore the effects of a range of confounding effects on metabolite levels. (2) Model individual or biologically related groups of metabolite levels using multivariate statistical approaches to determine the contribution of clinical/demographic and pre-analytical variables and their predictability across collection site. (3) Investigate the interactions between metabolites and clinical/demographic variables using machine learning approaches to identify stable metabolites and key interactions. (4) Provide the community with user-friendly software packages for the prediction of blood metabolite levels under homeostasis. An overall model of the metabolite concentrations in blood will be highly useful for a number of applications that include a better understanding of systems biology at the whole organism level, and ultimately improved risk prediction, disease diagnosis, treatment monitoring and outcomes analysis.
A quantitative model of blood homeostasis based on predicting the normal levels of small molecules in the blood can help identify diseases or other stresses that cause changes to those levels. The proposed statistical methods that will be used to develop this homeostasis model have the potential to efficiently identify more reliable disease markers and to more accurately predict disease risk.
