Inferring Disease-Related Metabolite Dependencies with a Bayesian Optimization Algorithm

• Published in: Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics pp. 62 - 73
• Main Authors: Franken, Holger, Seitz, Alexander, Lehmann, Rainer, Häring, Hans-Ulrich, Stefan, Norbert, Zell, Andreas
• Format: Book Chapter
• Language: English
• Published: Berlin, Heidelberg Springer Berlin Heidelberg 2012
• Series: Lecture Notes in Computer Science
• Subjects: Bayesian Information Criterion; Bayesian Network; Feature Selection; Feature Subset; Nonalcoholic Fatty Liver Disease
• ISBN: 9783642290657; 3642290655
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-642-29066-4_6

Understanding disease-related metabolite interactions is a key issue in computational biology. We apply a modified Bayesian Optimization Algorithm to targeted metabolomics data from plasma samples of insulin-sensitive and -resistant subjects both suffering from non-alcoholic fatty liver disease. In addition to improving the classification accuracy by selecting relevant features, we extract the information that led to their selection and reconstruct networks from detected feature dependencies. We compare the influence of a variety of classifiers and different scoring metrics and examine whether the reconstructed networks represent physiological metabolite interconnections. We find that the presented method is capable of significantly improving the classification accuracy of otherwise hardly classifiable metabolomics data and that the detected metabolite dependencies can be mapped to physiological pathways, which in turn were affirmed by literature from the domain.