Fig. 5: Thor provides unbiased screening of hallmarks in cancer.

a H&E staining image of the DCIS tissue. The annotation of eighteen major tumor regions (T1–T18) in the DCIS tissue is adapted from the annotation by pathology experts (Agoko NV, Belgium). b Leiden clusters of the segmented cells using morphological features. The list of image features and details of Leiden clustering are provided in Supplementary Note 1. Colors represent cell clusters. c The spatial distribution of cell types. Cell types are obtained by Cell-ID using the Thor-inferred spatial transcriptome of the in silico cells and refined with cell type markers. d VEGFA gene expression pattern at tissue and cell scales in tumor region T1. The boxes mark regions of interest. e The tumor regions identified by high attention values in CLAM and semi-supervised annotation in Mjolnir. The solid black boxes mark the curated regions for semi-supervised annotation. The dotted black square marks a high-attention region where adipocytes are predominantly located. f Heatmap of the copy number profiles inferred by CopyKAT based on the in silico cell-level transcriptome predicted by Thor. Representative breast cancer-related genes are provided. g Aneuploid (tumor) and diploid (non-tumor) regions inferred by CopyKAT show consistent results between the in silico cell-level transcriptome and the spot data. Source data are provided in a Source Data file.