Fig. 2

Screening important genes related to the prognosis of AML patients based on LASSO and RF models. (A) and (B) indicate that LASSO (least absolute shrinkage and selection operator) screened 20 important genes associated with AML prognosis. The method uses an L1 penalty to shrink some regression coefficients to exactly zero. (A) Ten time cross-validation for tuning parameter selection in the LASSO model; The binomial deviance curve was plotted versus log (λ), where λ is the tuning parameter. (B) LASSO coefficient profiles; LASSO coefficient profiles of clinic pathologic variables. (C), (D) and (E) indicate that the RF (random forest) algorithm screened the top five genes ranked by importance, which were related to AML prognosis. (C) The effect of the number of decision trees on the error rate (when the number of decision trees is about 2000, the error rate is relatively stable); The x-axis represents the number of decision trees and the y-axis represents the error rate. (D) Gini coefficient method in random forest classifier. x-axis: the genetic variable; y-axis: the importance index. (E) The ROC curve of the RF model, The AUC (area under the ROC curve) value is 0.977, which indicates that the predictive performance of the RF model is good