Figure 2 | Scientific Reports

Figure 2

From: Efficient prediction of a spatial transcriptomics profile better characterizes breast cancer tissue sections without costly experimentation

Figure 2

DeepSpaCE predicts spatial gene expression and cluster types. (a) Left image shows section D2 after hematoxylin and eosin (H&E) staining. Middle image shows a heatmap of normalized ESR1 expression in section D2, measured using the 10x Genomics Visium platform. ESR1 expression in the upper right region (black arrow) of section D2 could not be measured because of potential permeabilization errors. Right image shows the heatmap of ESR1 expression in section D2, predicted by DeepSpaCE. The blank areas represent spots excluded because of a small amount of information. (b) Image showing immunohistochemical staining of ESR1 protein in the adjacent section of section D2. (c) Heatmap shows the confusion matrix of the measured and predicted cluster types. The number in each box is the number of labels, and the number inside each pair of parentheses is the recall value. (d) Left image shows a heatmap of transcriptomic cluster types in section D2, as measured using the Visium platform. The cluster types in the upper right region (black arrow) of section D2 were not determined because of potential permeabilization errors. Right image shows the heatmap of cluster types in section D2, predicted by DeepSpaCE. Blank areas represent spots that were excluded because of low information. For training and gene prediction parts, we excluded cluster 8 because most of the region belonging to the cluster showed potential permeabilization errors.

Back to article page