Fig. 6: SpotClean improves identification of malignant spots in human breast cancer sample human_breast_2.

a Malignant spot composition as estimated via SPOTlight²² is shown for the raw data (upper left) and SpotClean decontaminated data (upper middle). Scale bar is 1 mm. The raw data identifies many spots as malignant whereas the SpotClean decontaminated data more closely resembles the annotations derived from the H&E image (upper right). The insets highlighted in the upper panel are shown in the lower panel. b Spearman correlations between average expression in the malignant scRNA-seq cells and spot-specific expression were calculated. Boxplots of correlations are shown for n = 265 strongly non-malignant spots, 216 questionably malignant spots (spots labeled malignant in the raw data, but not the SpotClean decontaminated data), and 546 strongly malignant spots. The lower whisker, lower hinge, line inside box, upper hinge, and upper whisker represent the minimum, lower quartile, median, upper quartile, and maximum calculated without outlier values which are more than 1.5*inter-quartile range away from the hinges and are shown in separate dots. Correlations with non-malignant scRNA-seq cells are also shown. The correlations show that expression in the questionably malignant spots more closely resembles that in non-malignant cells suggesting that the malignant classification in the raw data at these spots is likely false due to spot swapping. c The UMAP plot further demonstrates that the questionably malignant spots in the raw data are likely false positives as their expression more closely resembles that at non-malignant spots.