A urine metabolomics study based on gas chromatography-mass spectrometry (GC-MS) and multivariate statistical analysis was applied to distinguish rat bladder cancer. Urine samples with different stages were collected from animal models, i.e., the early stage, medium stage, and advanced stage of the bladder cancer model group and healthy group. After resolving urea with urease, the urine samples were extracted with methanol and, then, derived with N, O-Bis(trimethylsilyl) trifluoroacetamide and trimethylchlorosilane (BSTFA + TMCS, 99 : 1, v/v), before analyzed by GC-MS. Three classification models, i.e., healthy control vs. early- and middle-stage groups, healthy control vs. advanced-stage group, and early- and middle-stage groups vs. advanced-stage group, were established to analyze these experimental data by using Random Forests (RF) algorithm, respectively. The classification results showed that combining random forest algorithm with metabolites characters, the differences caused by the progress of disease could be effectively exhibited. Our results showed that glyceric acid, 2, 3-dihydroxybutanoic acid, N-(oxohexyl)-glycine, and D-turanose had higher contributions in classification of different groups. The pathway analysis results showed that these metabolites had relationships with starch and sucrose, glycine, serine, threonine, and galactose metabolism. Our study results suggested that urine metabolomics was an effective approach for disease diagnosis.
An output-based knowledge transfer approach and its application in bladder cancer prediction
