Fig. 4

Feature elimination curves of hub genes and heatmap of the 12 optimal feature genes in different dataset. A Feature elimination curves of hub genes. Root mean square error (RMSE) is the statistical parameter to determine the optimal feature genes after the analysis of recursive feature elimination algorithm. The lowest RMSE correspond with the best optimal feature gene set, based on which the model was trained by machine learning methods in 50% samples in GSE12288. B–E Heatmap of the 12 optimal feature genes in different dataset using pheatmap package. The red and blue colors indicate high and low expression, respectively, of the 12 optimal feature genes among samples. Upregulation, genes that higher expressed in case group than control group. Downregulation, genes that lower expressed in case group than control group. Integrated data was the combination of GSE12288, GSE7638 and GSE66360