Rapid Prediction of Bacterial Heterotrophic Fluxomics Using Machine Learning and Constraint Programming

• DOI: 10.1371/journal.pcbi.1004838
• Publication Date: 2016-04-19T18:19:49Z
• Authors: Stephen Gang Wu; Yuxuan Wang; Wu Jiang; Tolutola Oyetunde; Ruilian Yao; Xuehong Zhang; Kazuyuki Shimizu; Yinjie J. Tang; Forrest Sheng Bao
• Keywords: Metabolic flux analysis

13C metabolic flux analysis (13C-MFA) has been widely used to measure in vivo enzyme reaction rates (i.e., flux) microorganisms. Mining the relationship between environmental and genetic factors fluxes hidden existing fluxomic data will lead predictive models that can significantly accelerate quantification. In this paper, we present a web-based platform MFlux (http://mflux.org) predicts bacterial central metabolism via machine learning, leveraging from approximately 100 13C-MFA papers on heterotrophic metabolisms. Three learning methods, namely Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), Decision Tree, were employed study sophisticated influential fluxes. We performed grid search of best parameter set for each algorithm verified their performance through 10-fold cross validations. SVM yields highest accuracy among all three algorithms. Further, quadratic programming adjust profiles satisfy stoichiometric constraints. Multiple case studies have shown reasonably predict fluxomes as function species, substrate types, growth rate, oxygen conditions, cultivation methods. Due interest studying model organism under particular carbon sources, bias fluxome dataset may limit applicability models. This problem be resolved after more are published non-model species.