Metabolite marker discovery for the detection of bladder cancer by comparative metabolomics

// Chi-Hung Shao 1, * , Chien-Lun Chen 2, 3, * , Jia-You Lin 4 , Chao-Jung Chen 5, 6, 7 , Shu-Hsuan Fu 8 , Yi-Ting Chen 8, 9 , Yu-Sun Chang 8 , Jau-Song Yu 8, 9, 10 , Ke-Hung Tsui 2 , Chiun-Gung Juo 8 and Kun-Pin Wu 1 1 Institute of Biomedical Informatics, National Yang Ming University, Taipei 11221, Taiwan 2 Department of Urology, Chang Gung Memorial Hospital, Taoyuan 33305, Taiwan 3 College of Medicine, Chang Gung University, Taoyuan 33302, Taiwan 4 Department of Biomedical Engineering, National Yang Ming University, Taipei 11221, Taiwan 5 Proteomics Core Laboratory, China Medical University Hospital, Taichung 40402, Taiwan 6 Department of Medical Research, China Medical University Hospital, Taichung 40402, Taiwan 7 Graduate Institute of Integrated Medicine, China Medical University, Taichung 40402, Taiwan 8 Molecular Medicine Research Center, Chang Gung University, Taoyuan 33302, Taiwan 9 Department of Biomedical Sciences, Chang Gung University, Taoyuan 33302, Taiwan 10 Department of Cell and Molecular Biology, Chang Gung University, Taoyuan 33302, Taiwan * These authors have contributed equally to this work Correspondence to: Kun-Pin Wu, email: kpwu@ym.edu.tw Chiun-Gung Juo, email: cgjuomail@gmail.com Keywords: bladder cancer, metabolomics, metabolite marker selection, decision tree, machine learning Received: June 29, 2016     Accepted: February 28, 2017     Published: March 21, 2017 ABSTRACT Bladder cancer is one of the most common urinary tract carcinomas in the world. Urine metabolomics is a promising approach for bladder cancer detection and marker discovery since urine is in direct contact with bladder epithelia cells; metabolites released from bladder cancer cells may be enriched in urine samples. In this study, we applied ultra-performance liquid chromatography time-of-flight mass spectrometry to profile metabolite profiles of 87 samples from bladder cancer patients and 65 samples from hernia patients. An OPLS-DA classification revealed that bladder cancer samples can be discriminated from hernia samples based on the profiles. A marker discovery pipeline selected six putative markers from the metabolomic profiles. An LLE clustering demonstrated the discriminative power of the chosen marker candidates. Two of the six markers were identified as imidazoleacetic acid whose relation to bladder cancer has certain degree of supporting evidence. A machine learning model, decision trees, was built based on the metabolomic profiles and the six marker candidates. The decision tree obtained an accuracy of 76.60%, a sensitivity of 71.88%, and a specificity of 86.67% from an independent test.