Fig. 3.

Screening m⁷G regulator diagnostic markers by three feature selection algorithms. A Bayesian information criterion score by feature inclusion of best subset regression (BSR) analysis. B Model performance based on different feature subsets in BSR analysis. C Least absolute shrinkage and selection operator (LASSO) regression algorithm to identify diagnostic markers. D RIDGE regression algorithm to identify diagnostic markers. E Elastic net (EN) regression algorithm to identify diagnostic markers for HF. F Root mean squared error (RMSE) of three regularization technique models in the internal validation dataset. G Out-of-bag (OOB) error rate of the random forest (RF) model. H Search for the optimal value of mtry for RF model. I Variable importance plot for the RF model. The features are ranked by the mean decrease in classification accuracy when they are permuted. The more the Gini coefficient decreases on average, the more important the variable is. J Venn diagram showing the intersected genes of BSR analysis, RIDGE regression and RF algorithm