Fig 1.
TUSCAN is designed to identify tumor regions in spatially resolved transcriptomics data. First, it integrates the histology image and spatial gene expression data as inputs; it then employs an autoencoder for dimensionality reduction. The low-dimensional features serve as the basis for an initial clustering of all spots. The cluster with the highest confidence of representing normal cells is defined (left box). A CNP for all spots is constructed (middle box) with the selected cluster as a normal reference. Finally, TUSCAN segments all spots into tumor regions and normal regions via consensus clustering with the CNP as the input. TUSCAN is also capable of performing a tumor subclone analysis, reconstructing a tumor clonal tree, and performing a regional differential analysis (right box). Created in BioRender. Zang, C. (2026) https://BioRender.com/gtpfhy9.
Fig 2.
Benchmark evaluation and comparison with other methods.
Pathologist-annotated histology images are in the left panel. Tumor sections of four human cancer samples identified by TUSCAN, TESLA, BayesSpace, CopyKat, SpatialInferCNV, and SCEVAN are marked in red. (A) Human breast cancer (Ductal Carcinoma in Situ). (B) Human breast cancer (Block A, Section 1). (C) Human prostate cancer. (D) Human epidermal growth factor receptor 2 + breast cancer, patient B.
Fig 3.
Tumor clone analysis of human breast cancer.
(A) Spatial visualization of three tumor clones and normal tissue sections. (B) Heatmap of the estimated CNPs of tumor spots. Clonal classifications were determined by consensus clustering. (C) CNV burden calculated as the number of genes with extreme expression levels relative to the normal group. CNV_gain: . CNV_loss:
. (D) Spatially variable features across three tumor clones. Top to bottom: clone 1 vs. clones 2 and 3; clone 2 vs. clones 1 and 3; clone 3 vs. clones 1 and 2. (E) Differential expression analysis of three tumor clones and normal cells. (F) GSEA of three tumor clones. Human hallmark gene sets from the Molecular Signatures Database (MSigDB) were used as the reference.
Fig 4.
Further subclone tree reconstruction.
(A) Spatial visualization of six tumor clones. (B) Heatmap of estimated CNPs for tumor spots. Clone 3 was further subdivided into four subclones. (C) Phylogenetic clonal tree of the tumor clones from inferred CNPs. Gray circles are unknown ancestors. The pink circle represents the original normal cells. All CNV locations of clone C are marked on the left side of the node, where the maximum CNV locations are observed. CNV locations that are unique to other clones in comparison to clone C are indicated with distinct colors. The complete CNV profile can be found in the S3 Table. (D) Visualization of the inferred trajectory. Higher pseudo-times are marked as darker colors.