Journal: Computational and Structural Biotechnology Journal

Article Title: A comprehensive overview of graph neural network-based approaches to clustering for spatial transcriptomics

doi: 10.1016/j.csbj.2023.11.055

Figure Lengend Snippet: This review considers the task of spatial clustering on eight spatially resolved transcriptomics datasets. A. The human DLPFC dataset consisted of 12 neuronal samples from three subjects, each with four sample sections. The displayed sample ID is 151510. B. The mouse brain dataset was obtained from 10X Visium, and each spot contained more than one cell. The entire dataset and its cropped section are available in the Squidpy package. C. The human breast cancer dataset was downloaded from 10X Visium, with manual annotations acquired from the SEDR package. D. The dataset of the olfactory bulbs of mouse was derived from Slide-seqV2 technology, and the particular section Puck_200127_15 was used. E. The Stereo-seq-based dataset of the olfactory bulbs of mouse was obtained from the original paper, and the special section was obtained from the results of SEDR analysis. F. The mouse liver dataset obtained from the Vizgen MERFISH Mouse Liver Map. The special slice is the Liver1Slice1 that measures 347 genes across over > 300,000 liver cells in a single mouse liver slice. G. The pre-processed subset of seqFISH data acquired from Squidpy package. H. The 10x Xenium dataset from the Xenium breast cancer tumor microenvironment Datbase. The related pre-processed data can be gained from Squidpy as well.

Article Snippet: Finally, the latent representations are obtained in an end-to-end manner and are refined by using a deep clustering method to ensure their compactness. conST has been evaluated on the mouse hypothalamus (MERFISH), mouse visual cortex (seqFISH), human DLPFC (10x Visium), human breast cancer (10x genomics), and datasets of the olfactory bulbs of mice (Stereo-seq). conST has also been compared with benchmarking methods of clustering, including Seurat, Giotto, stLearn, SpaGCN, SEDR, and BayesSpace.

Techniques: Derivative Assay

Journal: Computational and Structural Biotechnology Journal

Article Title: A comprehensive overview of graph neural network-based approaches to clustering for spatial transcriptomics

doi: 10.1016/j.csbj.2023.11.055

Figure Lengend Snippet: Clustering performance of eight benchmarking methods on the human DLPFC dataset. A. The mean adjusted Rand index (ARI) of each clustering method was determined (Upper). The variance of each clustering method. GraphST had the highest average ARI and the lowest variance (Middle). The ARI of each spatial clustering algorithm was calculated on 12 samples of the human DLPFC. The eight methods are ranked from 1 to 8 in order from the highest to the lowest ARI (Lower). B. The Normalized Mutual Information (NMI) of each clustering method was determined (Upper). The variance of each clustering method. Spatial-MGCN had the highest average NMI and the lowest variance (Middle). The NMI of each spatial clustering algorithm was calculated on 12 samples of the human DLPFC. The eight methods are ranked from 1 to 8 in order from the highest to the lowest NMI (Lower).

Article Snippet: Finally, the latent representations are obtained in an end-to-end manner and are refined by using a deep clustering method to ensure their compactness. conST has been evaluated on the mouse hypothalamus (MERFISH), mouse visual cortex (seqFISH), human DLPFC (10x Visium), human breast cancer (10x genomics), and datasets of the olfactory bulbs of mice (Stereo-seq). conST has also been compared with benchmarking methods of clustering, including Seurat, Giotto, stLearn, SpaGCN, SEDR, and BayesSpace.

Techniques:

Journal: Computational and Structural Biotechnology Journal

Article Title: A comprehensive overview of graph neural network-based approaches to clustering for spatial transcriptomics

doi: 10.1016/j.csbj.2023.11.055

Figure Lengend Snippet: The ARI and NMI scores of eight spatial clustering methods on three selected DLPFC samples. A, B, C. Three samples were obtained from three different subjects participating in the human DLPFC project. Each subject provided four sections of human brain tissues, and we selected one section from each to compare their ARI values and NMI. Samples 151510 and 151676 were annotated manually into seven clusters, while sample 151672 contained only five groups in the ground truth. The black segment visually represents the standard deviation.

Article Snippet: Finally, the latent representations are obtained in an end-to-end manner and are refined by using a deep clustering method to ensure their compactness. conST has been evaluated on the mouse hypothalamus (MERFISH), mouse visual cortex (seqFISH), human DLPFC (10x Visium), human breast cancer (10x genomics), and datasets of the olfactory bulbs of mice (Stereo-seq). conST has also been compared with benchmarking methods of clustering, including Seurat, Giotto, stLearn, SpaGCN, SEDR, and BayesSpace.

Techniques: Standard Deviation

Journal: Computational and Structural Biotechnology Journal

Article Title: A comprehensive overview of graph neural network-based approaches to clustering for spatial transcriptomics

doi: 10.1016/j.csbj.2023.11.055

Figure Lengend Snippet: The identified spatial domains of eight compared methods on three selected slices. A, B, C. The cell-type clustering task was performed on three samples (151570, 151672, and 151676) by using eight spatial clustering methods: SpaGCN, conST, DeepST, CCST, STAGATE, GraphST, SpaceFlow and Spatial-MGCN. These samples corresponded to the three subjects in the human DLPFC dataset. A comparison was made with the ground truth to assess the clustering performance of each baseline.

Article Snippet: Finally, the latent representations are obtained in an end-to-end manner and are refined by using a deep clustering method to ensure their compactness. conST has been evaluated on the mouse hypothalamus (MERFISH), mouse visual cortex (seqFISH), human DLPFC (10x Visium), human breast cancer (10x genomics), and datasets of the olfactory bulbs of mice (Stereo-seq). conST has also been compared with benchmarking methods of clustering, including Seurat, Giotto, stLearn, SpaGCN, SEDR, and BayesSpace.

Techniques: Comparison

Journal: Briefings in Bioinformatics

Article Title: CHAI: consensus clustering through similarity matrix integration for cell-type identification

doi: 10.1093/bib/bbae411

Figure Lengend Snippet: CHAI-ST benchmarking on human DLPFC 10X visium datasets and Savas breast cancer dataset.

Article Snippet: Since the benchmarking results in show that integrating the spatial transcriptomic results into CHAI-ST-SNF at the second level yielded the best results, we chose to use this method for our evaluation, in addition to CHAI-AvgSim-ST. Sicnce STGNNks relies on 10X Genomics Visium datasets as input, we compared both CHAI-ST methods to the baseline methods on three human DLPFC 10X Visium datasets [ , ].

Techniques:

Journal: Briefings in Bioinformatics

Article Title: Deconvolution of spatial transcriptomics data via graph contrastive learning and partial least square regression

doi: 10.1093/bib/bbaf052

Figure Lengend Snippet: Summary of the spatial transcriptomics and single cell datasets.

Article Snippet: We further applied CLPLS to the ST data of the human dorsolateral prefrontal cortex (DLPFC) obtained by 10X Visium and scRNA-seq data obtained by 10X Genomics Chromium [ , ].

Techniques: Biomarker Discovery

Journal: Briefings in Bioinformatics

Article Title: Deconvolution of spatial transcriptomics data via graph contrastive learning and partial least square regression

doi: 10.1093/bib/bbaf052

Figure Lengend Snippet: Summary of the spatial transcriptomics and single cell datasets.

Article Snippet: We further applied CLPLS to the ST data of the human dorsolateral prefrontal cortex (DLPFC) obtained by 10X Visium and scRNA-seq data obtained by 10X Genomics Chromium [ , ].

Techniques: Biomarker Discovery