A Knowledge-Driven Network-Based Analytical Framework for the Identification of Rumen Metabolites

Metabolites are the final production of biochemical reactions in the rumen micro-ecological system and are very sensitive to changes in rumen microbes. Nuclear magnetic resonance (NMR) spectroscopy could both identify and quantify the metabolic composition of the ruminal fluid, which reflects the interaction between rumen microbes and diet. The main challenge of untargeted metabolomics is the compound annotation. Based on non-linear and linear associations between microbial gene abundances and integrals derived from NMR spectra, combined with knowledge of enzymatic reaction from the KEGG database, this study developed a knowledge-driven network-based analytical framework for the inference of metabolites. There were 89 potential metabolites inferred from the integral co-occurrence network. The results are supported by dissimilarity network analysis. The coexistence of non-linear and linear associations between microbial gene abundances and spectral integrals was detected. The study successfully found the corresponding integrals for acetate, butyrate and propionate, which are the major volatile fatty acids (VFA) in the rumen. This novel framework could very efficiently infer metabolites to corresponding integrals from NMR spectra.