A novel network modelling for metabolite set analysis: a case study on crc metabolomics

• Main Authors: Liu, Y., Xu, X., Deng, L., Cheng, K. K., Xu, J., Raftery, D., Dong, J.
• Format: Article
• Published: Institute of Electrical and Electronics Engineers Inc. 2020
• Subjects: TP Chemical technology
• Online Access: http://eprints.utm.my/id/eprint/87026/; http://www.dx.doi.org/10.1109/ACCESS.2020.3000432

In metabolomics, pathway analysis normally refers to analysis of a pre-defined sets of metabolites (metabolite set) associated to the metabolic pathways. The metabolite set analysis is useful to facilitate biological interpretation of metabolomics data. The currently available methods may be divided into three generations: over-representation analysis, functional class scoring, and network topology analysis. Among the three generations of tools, the network topology methods have been shown to have lower false discovery rates and better biological interpretability than the other two earlier generations of tools. However, most of the current network topology methods focus the analysis only at the metabolite-level network. The interaction between pathways are not taken into consideration. To address this issue, we propose a new metabolite sets association network (MSAN) modelling scheme. In the developed method, the metabolite sets are defined based on the KEGG databases. By using the metabolite sets as vertexes, the MSAN network evaluated the relationships between pairs of metabolite sets based on their mutual information. The impact of a single metabolite set on the overall network was evaluated by the MSAN network, which may help to uncover differential metabolite sets relevant to the underlying biology mechanism of the study. A metabolomic dataset from a published colorectal cancer (CRC) study is used to evaluate the performance of MSAN network to identify perturbed metabolite sets in colorectal cancer patients. The current results are compared to that of two commonly used methods, NetGSA and MetaboAnalyst, which are based on the metabolite-level network approach. The current method highlights a number of metabolites sets consistent with recent published CRC reports. Taken together, the proposed method may provide an alternative tool for the identification of dysregulated pathways and facilitate biological interpretation of metabolomics data.