Fig. 6
From: An uncertainty-based interpretable deep learning framework for predicting breast cancer outcome
Breast cancer-related gene selection results produced by different methods. a Top 200 genes selected by computing the importance of the features in UISNet. The x-axis represents the average importance values of the different genes, and the y-axis is the uncertainty weight value computed by Eq. (12). b The results were used to identify the top 200 genes ranked based on the |log(fold change)| of DEA (adjusted-p-values < 0.05). c Heatmap of the 200 identified differentially expressed genes. d Breast cancer outcome prediction performance achieved by using different numbers of selected gene features based on DEA, the IG-based method without the uncertainty strategy (IG), and UISNet. A Venn diagram was used to show the number of overlapping genes selected by different methods