Fig 1.
SpaConTDS facilitates accurate spatial domain identification and downstream tasks in HER2+ and DLPFC.
(A) H&E stained image and manual annotations of HER2+ section D1. (B) Boxplots of ARI and NMI of the twelve methods applied to all 7 HER2+ sections. (C) Clustering results with ARI of SpaConTDS and baseline methods on HER2+ section D1. (D) H&E stained image and manual annotations of DLPFC slice 151671. (E) Boxplots of ARI and NMI of the twelve methods applied to all 12 DLPFC slices. (F) Clustering results with ARI of SpaConTDS and baseline methods on DLPFC slice 151671. (G) UMAP visualization of SpaConTDS and baseline methods on DLPFC slice 151671. (H) PAGA trajectory graphs of SpaConTDS and baseline methods on DLPFC slice 151671.
Fig 2.
Advanced spatial analysis by SpaConTDS reveals detailed spatial stratification for human breast cancer.
(A) H&E stained image, manual annotations and clustering result with ARI of SpaConTDS on human breast cancer dataset. (B) Violin plots of expression of DEGs (EFHD1, SHISA2, CTTN) in subcluster 2 and 15 versus other clusters. (C) Clustering results with ARI of baseline methods on human breast cancer dataset. (D) H&E stained image with manually annotated regions and clustering result with DB index of SpaConTDS on IDC dataset. (E) Top: raw gene expression patterns of CXCL14, MUC1, TCEAL4andCCND1. Bottom: denoised gene expression patterns of CXCL14, MUC1, TCEAL4andCCND1. (F) Clustering results with DB index of baseline methods on IDC dataset.
Fig 3.
Integration abilities of SpaConTDS.
(A) Slice sampling diagram and manual annotations of DLPFC slice 151673, 151674, 151675 and 151676. (B) Overlapping diagram of three partially overlapping slices from the human placental bed dataset. (C) Clustering results with ARI of SpaConTDS, STAGATE, GraphST, Scanpy and IRIS on four DLPFC slices. (D) Vertical integration results of DLPFC slices. UMAP plots after batch effect correction (top) and spatial clustering results(bottom) from SpaConTDS, STAGATE, GraphST, Scanpy and IRIS. (E) Partial overlap integration results of human placental bed dataset. UMAP plots after batch effect correction (top) and spatial clustering results (bottom) from SpaConTDS, STAGATE, GraphST, Scanpy and IRIS.
Fig 4.
SpaConTDS achieves superior clustering performance on breast cancer 10x Xenium data.
(A) H&E stained image with manually annotated regions and clustering result with DB index of SpaConTDS on 10x Xenium breast cancer dataset. (B) Violin plots of expression of DEGs (SERPINA3, PTN) in subcluster 4 and 11 versus other clusters. (C) Raw(top) and denoised(bottom) gene expression patterns of PTN. (D) Clustering results with DB index of STAGATE, ConST, GraphST, Scanpy and IRIS.
Fig 5.
SpaConTDS enhances resolution of interface dynamics in zebrafish melanoma.
(A) H&E stained image and manual annotations of zebrafish melanoma on slices A and B from [3] with no changes made, and this article is licensed under a Creative Commons Attribution 4.0 international license. (B) Interface domains identified by SpaConTDS on slices A and B with DB index. (C) The expression of select marker genes(muscle(top): ATP2A1L, CKMA, AK1; tumor(bottom) : BRAFhuman, HMGB2A, and B2M).
Fig 6.
SpaConTDS improves spatial transcriptomic analysis through each module.
(A) H&E stained image and manual annotations of DLPFC slice 151673. (B) Clustering results with ARI based purely on image features by SpaConTDS and stLearn on DLPFC slice 151673. (C) H&E stained image and manual annotations of IDC dataset. (D) Clustering results with DB based purely on image features by SpaConTDS and stLearn on IDC dataset. (E) Boxplot of ARI using SpaConTDS and SpaConTDS without each module on all 7 HER2+ sections.
Fig 7.
Schematic overview of SpaConTDS.
(A) Framework of SpaConTDS. (B) Overview of the positive and negative samples construction strategy. (C) For multiple tissue slices, SpaConTDS constructs a global similarity matrix across all slices to generate positive samples with cross-slice information. (D) Downstream tasks.