Fig. 4

Investigation of metabolic markers for lung cancer detection and screening. (a) The OPLS-DA score plot model for differentiation of early lung cancer and healthy control. (b) Volcano plot showing expression levels of differential metabolites. (c) The top ten metabolites with the largest mean decrease accuracy value and mean decrease Gini value in the random-forest model. (d) The ROC curve was derived from the training set, testing set, and predicting set using four metabolites selected by the random-forest model, respectively. (e-g) Heat map showing the expression level of four metabolic markers in early lung cancer and control samples, including training set, testing set, and predicting set. “+” represents lung cancer patient, and “-” represents control.