A novel strategy for rapidly and accurately screening biomarkers based on ultraperformance liquid chromatography-mass spectrometry metabolomics data.

We reported a novel strategy for rapidly and accurately screening biomarkers based on ultraperformance liquid chromatography-mass spectrometry (UPLC-MS) metabolomics data. First, the preliminary variables were obtained by screening the original variables using method validation. Second, the variables were selected from the preliminary variables and formed the variable sets by testing different thresholds of single factor (variable importance in projection (VIP), fold change (FC), the area under the receiver operating characteristic curve (AUROC), and -ln(p-value)). Then the partial least squares-discriminant analysis (PLS-DA) models were performed. The best threshold of each factor, and the corresponding variable set were found by comparing the models' R2X, R2Y, and Q2. Third, the second-step-obtained variable sets were further screened by multi-factors. The best combination of the multi-factors, and the corresponding variable set were found by comparing R2X, R2Y, and Q2. The expected biomarkers were thus obtained. The proposed strategy was successfully applied to screen biomarkers in urine, plasma, hippocampus, and cortex samples of Alzheimer's disease (AD) model, and significantly decreased the time of screening and identifying biomarkers, improved the R2X, R2Y, and Q2, therefore enhanced the interpreting, grouping, and predicting abilities of the PLS-DA model compared with generally-applied procedure. This work can provide a valuable clue to scientists who search for potential biomarkers. It is expected that the developed strategy can be written as a program and applied to screen biomarkers rapidly, efficiently and accurately.