|
Fig. 5 Machine learning analysis reveals EZH2 as a key gene in oxidative stress-regulated circadian rhythm (A) After multiple GEO datasets were merged, batch effects were removed via the Combat package, with boxplots showing data distributions before and after batch effect removal. (B–C) Volcano plots and heatmaps displaying the distribution of upregulated and downregulated genes in the dataset. (D–E) KEGG and GO enrichment analysis results of differentially expressed genes (DEGs). (F) GSEA enrichment analysis using the Misdb circadian gene set as the background. (G) WGCNA weighted gene coexpression network analysis divided the dataset into multiple gene modules, with the black module showing the highest correlation to the circadian clock. (H) Venn diagram showing the intersection of WGCNA module genes and DEGs. (I–J) Residual plots from the RF, LASSO, and SVM machine learning methods revealing the reliability of the three methods. (K) Gene importance ranking based on machine learning models. (L) Box-and-scatter plots of TPM values of the EZH2 gene across different groups in the dataset, with a t-test (∗∗∗, p < 0.001) indicating significant differences between the two groups.