Extended Data Fig. 3: Clustering results for a human breast cancer spatial transcriptomics dataset.

a, Pathologist manual annotation of tissue section. DCIS: Ductal carcinoma in situ. b, Shown from left to right are clustering results from MISO, MUSE, and SpatialGlue, respectively. Two patches were selected to highlight that MISO’s results agree better with histological patterns. c, RNA and image ICC distributions across all clusters and features for each method in the breast cancer data (n = 750 ICC values for each group). The mean ICC for each method and each modality is printed on the corresponding box plot. Test statistics and p-values were obtained using one-sided (<,>) or two-sided (≈) t-tests. d, Spots plotted according to their RNA t-SNE coordinates and colored by the clustering results for each method. The MISO clustering results demonstrate coherence with respect to gene expression patterns and the annotated histological regions. e, RNA t-SNE plot for the breast cancer Visium dataset with spots colored according to total UMI count. MISO was able to localize a sub-cluster in the annotated invasive carcinoma region with much lower total UMI counts compared to other sub-clusters in this region. f, SpatialGlue clustering results when increasing the weight given to histology in the loss function. SpatialGlue was not able to detect the fat region of the tissue section when making the weight given to histology 10 or 50 times greater than that given to gene expression. Box plots: center line, median; box limits, upper and lower quartiles; whiskers, 1.5x interquartile range; points, outliers.