Fig. 3

Machine learning to determine risk score for discriminating high-and low-risk stage IA lung adenocarcinoma. A Risk-related genes by logistic regression (high-risk, H vs. low-risk, L) in each training set (k[1: k-1]; k = 1,2,3,4,5); B Prediction effect (classifying H and L) of risk-related gene (1: k-1) in test set (k) by five machine learning algorithms (left) and six ensemble models (right); C Intersection of risk-related genes in five trials: OR > 1 (left) and OR < 1 (right); D LASSO regression to filtrate intersection genes for establish risk score (IA score) for 386 samples (11 datasets not including GSE30219); E Prediction effect of IA score (classifying H and L) in training set (11 datasets not including GSE30219) (left) and external validation set (GSE30219) (right); F Relative weight of IA score and classical demographic factors (left) and driver mutations (right) for risk classifying; G Discrimination power of IA score and classical demographic factors (left) and driver mutations (right) for risk classifying in decision curve analysis. OR Odds Ratio, LASSO least absolute shrinkage and selection operator