Analyze metabolic data of African Americans using Bayesian network

Chang Li, The University of Texas School of Public Health


Background: The development of metabolomic databases has enabled the study of constructing metabolic network. Our study took advantage of these datasets and the metabolic profile measured from Atherosclerosis Risk in Communities (ARIC) Study, to try to build the metabolic network, and find its conspicuous features. Method: Metabolic data was obtained from ARIC study, and was pre-processed in previous studies. Our data included 509 metabolites and 1833 data points. There were 308 known metabolites and 201 metabolites with unknown identity. Bootstrap re-sampling was repeated 60 times to the original data, and hill-climbing algorithm was applied to each of the bootstrapped dataset. Then an averaged network with only high confidence arcs was obtained, and analyzed. Maximum likelihood estimation was used to estimate the parameters given the network structure. Biological sensible relationships between metabolites were identified using previous studies and publicly available databases. Result: The averaged network showed that metabolites with the same super-pathway tended to cluster together while the unknown metabolites spread out across the entire network. Our averaged network included 455 nodes and 753 arcs. The average neighbor size for each node was 1.655, with the highest being 11. Arachidonate was identified to have the most neighbors, and the most children, so was selected as candidate metabolite for further discussion. Discussion: Our network learned was stringent, and only the arcs with highest confidence were included. Many arcs with less confidence were shown as missing in our network, so the explanation of independence should be cautious. There were some hypotheses generated based on previous studies and our network. For example the EPA/AA ratio could have indicative effect on certain cardiovascular heart diseases in the African-American population. Such features and hypotheses identified could be selected as future research topics.

Subject Area

African American Studies|Public health|Biomechanics

Recommended Citation

Li, Chang, "Analyze metabolic data of African Americans using Bayesian network" (2014). Texas Medical Center Dissertations (via ProQuest). AAI1569949.