Prostate tumour sections with hypoxic cores [Visium SD FFPE]
Data files
Aug 01, 2026 version files 805.32 MB
-
dryad_upload.tar.gz
805.30 MB
-
README.md
19.23 KB
Abstract
Prostate Visium SD FFPE data associated with:
Proteomics-constrained deconvolution reveals spatial cell-type programs in tumours
Accurately resolving cell-type mixtures in spatial transcriptomics remains chal- lenging, particularly in heterogeneous tumours where cell populations are inter- mixed and matched single-cell references may be unavailable or poorly aligned. Current deconvolution approaches either require high-quality scRNA-seq priors, suffer from scalability limitations, or lack interpretability. We introduce PISTA- CHIO, a proteomics-informed spatial transcriptomics deconvolution framework based on constrained non-negative matrix factorization with a negative-binomial likelihood. Rather than using probabilistic priors, PISTACHIO incorporates hard spatial cell-type constraints derived from Imaging Mass Cytometry, enforcing biologically grounded sparsity and explicit spatial feasibility of cell-type pres- ence. PISTACHIO improved recovery of spatial cell-type distributions compared with Cell2location and STdeconvolve across synthetic and tumour datasets. Our approach remains robust under cell-type assignment errors, maintaining high cor- relation with ground truth under moderate noise, and achieves fast runtime on standard hardware, enabling practical large-scale deployment.
Dataset DOI: 10.5061/dryad.0cfxpnwhc
Description of the data and file structure
Prostate Visium SD FFPE data associated with:
Proteomics-constrained deconvolution reveals spatial cell-type programs in tumours
Accurately resolving cell-type mixtures in spatial transcriptomics remains chal- lenging, particularly in heterogeneous tumours where cell populations are inter- mixed and matched single-cell references may be unavailable or poorly aligned. Current deconvolution approaches either require high-quality scRNA-seq priors, suffer from scalability limitations, or lack interpretability. We introduce PISTA- CHIO, a proteomics-informed spatial transcriptomics deconvolution framework based on constrained non-negative matrix factorization with a negative-binomial likelihood. Rather than using probabilistic priors, PISTACHIO incorporates hard spatial cell-type constraints derived from Imaging Mass Cytometry, enforcing biologically grounded sparsity and explicit spatial feasibility of cell-type pres- ence. PISTACHIO improved recovery of spatial cell-type distributions compared with Cell2location and STdeconvolve across synthetic and tumour datasets. Our approach remains robust under cell-type assignment errors, maintaining high cor- relation with ground truth under moderate noise, and achieves fast runtime on standard hardware, enabling practical large-scale deployment.
Files and variables
File: dryad_upload.tar.gz
Description: Archive of reduced outputs from 10x Space Ranger
Selected files from Space Ranger output included, excluding BAM and QC Report; gene expression data can be extracted from "anon/RBJA_ST0X/filtered_feature_bc_matrix/". Full-resolution images can be provided on request.
The Visium SD FFPE data from two prostate tumour samples included within the manuscript are within these outputs as RBJA_ST01 & RBJA_ST02 (labeled subdirectories).
Description of file outputs are described on the 10x website: https://www.10xgenomics.com/support/software/space-ranger/latest/analysis/outputs/output-overview
Space Ranger processed data
Overview
The anon/ directory contains processed 1sion data for two samples:
RBJA_ST01RBJA_ST02
Both sample directories have the same organisation and contain the principal outputs generated by the 10x Genomics spaceranger count pipeline. The outputs include feature-by-spot count matrices, spatial image-registration information, quality-control metrics and automated secondary analyses.
In the descriptions below, [SAMPLE] represents either RBJA_ST01 or RBJA_ST02.
The presence of probe_set.csv indicates that these data were generated using a probe-based Visium assay, such as the Visium FFPE or Visium CytAssist workflow.
Directory structure
anon/
├── RBJA_ST01/
└── RBJA_ST02/
anon/
Top-level anonymised data directory containing the processed Space Ranger outputs for all submitted samples.
anon/[SAMPLE]/
Complete submitted Space Ranger output directory for one Visium capture area. Spot barcodes are sample-specific within each directory.
Automated secondary analysis
anon/[SAMPLE]/analysis/
Contains automated secondary analyses performed by Space Ranger using the filtered, tissue-associated feature-by-spot count matrix.
These analyses include principal component analysis, graph-based and K-means clustering, cluster-associated differential expression, t-SNE and UMAP dimensionality reduction. They are exploratory outputs and do not represent manually curated tissue or cell-type annotations.
Clustering
anon/[SAMPLE]/analysis/clustering/
Contains alternative cluster assignments generated from the expression profiles of tissue-associated spots.
analysis/clustering/graphclust/clusters.csv
Cluster assignment for each tissue-associated spot produced by Space Ranger's graph-based clustering procedure.
The graph-based method constructs a nearest-neighbour graph in principal-component space and identifies groups of spots with similar expression profiles. The CSV contains one row per spot barcode and its assigned cluster number.
analysis/clustering/kmeans_2_clusters/clusters.csv
analysis/clustering/kmeans_3_clusters/clusters.csv
analysis/clustering/kmeans_4_clusters/clusters.csv
analysis/clustering/kmeans_5_clusters/clusters.csv
analysis/clustering/kmeans_6_clusters/clusters.csv
analysis/clustering/kmeans_7_clusters/clusters.csv
analysis/clustering/kmeans_8_clusters/clusters.csv
analysis/clustering/kmeans_9_clusters/clusters.csv
analysis/clustering/kmeans_10_clusters/clusters.csv
Alternative K-means cluster assignments generated using predetermined numbers of clusters from K = 2 to K = 10.
Each file contains the spot barcode and its assigned cluster. These files allow the data to be examined at several levels of clustering granularity; the cluster numbers are specific to each individual clustering solution and should not be compared directly between different values of K.
Differential expression
anon/[SAMPLE]/analysis/diffexp/
Contains differential-expression results corresponding to each clustering solution.
For each cluster, Space Ranger compares spots assigned to that cluster against all other tissue-associated spots. The resulting tables report cluster-specific expression summaries, including mean expression, log2 fold change and multiple-testing-adjusted statistical significance for each feature.
analysis/diffexp/graphclust/differential_expression.csv
Differential-expression results for the graph-based clustering solution.
For each feature and graph-derived cluster, the file describes whether the feature is enriched or depleted in that cluster relative to the remaining spots.
analysis/diffexp/kmeans_2_clusters/differential_expression.csv
analysis/diffexp/kmeans_3_clusters/differential_expression.csv
analysis/diffexp/kmeans_4_clusters/differential_expression.csv
analysis/diffexp/kmeans_5_clusters/differential_expression.csv
analysis/diffexp/kmeans_6_clusters/differential_expression.csv
analysis/diffexp/kmeans_7_clusters/differential_expression.csv
analysis/diffexp/kmeans_8_clusters/differential_expression.csv
analysis/diffexp/kmeans_9_clusters/differential_expression.csv
analysis/diffexp/kmeans_10_clusters/differential_expression.csv
Differential-expression results corresponding to each K-means clustering solution.
Each feature is tested separately for enrichment in each cluster relative to all other spots. Because the spot assignments differ between clustering solutions, the associated differential-expression results may also differ between values of K.
Principal component analysis
anon/[SAMPLE]/analysis/pca/10_components/
Contains the results of principal component analysis performed on the normalised expression profiles of tissue-associated spots.
Space Ranger retained ten principal components for subsequent clustering and dimensionality-reduction analyses.
analysis/pca/10_components/components.csv
Feature loadings for the ten principal components.
The values describe the contribution and direction of each feature in each principal component. Features with large positive or negative loadings contribute strongly to the corresponding component.
analysis/pca/10_components/dispersion.csv
Normalised dispersion calculated for each feature.
Dispersion measures expression variability after accounting for average expression and is used to identify features that are informative for dimensionality reduction.
analysis/pca/10_components/features_selected.csv
Identifiers of the highly variable features selected for the PCA calculation.
Features not selected for PCA remain available in the expression matrices but did not contribute directly to calculation of the principal-component representation.
analysis/pca/10_components/projection.csv
Principal-component coordinates for each tissue-associated spot.
Each row contains a spot barcode followed by its values for principal components PC1 through PC10. These coordinates form the reduced representation used by the automated clustering and visualisation procedures.
analysis/pca/10_components/variance.csv
Proportion of expression variance explained by each principal component.
This file can be used to assess how much of the overall variation is represented by each of the ten retained components.
t-SNE coordinates
anon/[SAMPLE]/analysis/tsne/2_components/projection.csv
Two-dimensional t-distributed stochastic neighbour embedding, or t-SNE, coordinates for each tissue-associated spot.
The coordinates provide a nonlinear visualisation of expression similarity. Spots located close together in t-SNE space generally have similar expression profiles, but the coordinates do not represent physical positions within the tissue.
UMAP coordinates
anon/[SAMPLE]/analysis/umap/2_components/projection.csv
Two-dimensional uniform manifold approximation and projection, or UMAP, coordinates for each tissue-associated spot.
UMAP provides a nonlinear visualisation of relationships between spot expression profiles. As with t-SNE, these are expression-space coordinates rather than spatial tissue coordinates.
Feature-by-spot count matrices
Space Ranger provides the count matrices in both Matrix Exchange and HDF5 formats. The two formats contain substantially the same count information but are intended for different software environments.
Counts represent deduplicated unique molecular identifier, or UMI, counts for each feature and spatial barcode.
Filtered matrix in Matrix Exchange format
anon/[SAMPLE]/filtered_feature_bc_matrix/
Sparse feature-by-barcode matrix containing only barcodes corresponding to spots identified as being under tissue.
This is normally the principal matrix used for downstream analysis of Visium data.
filtered_feature_bc_matrix/barcodes.tsv.gz
Gzip-compressed list of tissue-associated spatial barcodes.
The order of the barcodes corresponds to the columns of matrix.mtx.gz.
filtered_feature_bc_matrix/features.tsv.gz
Gzip-compressed feature annotation table.
The file normally contains the feature identifier, human-readable feature name and feature type. The order of the features corresponds to the rows of matrix.mtx.gz.
filtered_feature_bc_matrix/matrix.mtx.gz
Gzip-compressed sparse feature-by-barcode UMI count matrix in Matrix Exchange format.
Rows correspond to entries in features.tsv.gz, while columns correspond to entries in barcodes.tsv.gz.
Raw matrix in Matrix Exchange format
anon/[SAMPLE]/raw_feature_bc_matrix/
Sparse feature-by-barcode matrix containing all detected spatial barcodes, including barcodes assigned to spots outside the tissue area.
This matrix is useful when tissue detection needs to be reassessed or when background and off-tissue signal are to be examined.
raw_feature_bc_matrix/barcodes.tsv.gz
Gzip-compressed list of all spatial barcodes represented in the raw matrix.
The barcode order corresponds to the columns of the associated matrix.mtx.gz file.
raw_feature_bc_matrix/features.tsv.gz
Gzip-compressed feature annotation table for the raw matrix.
The feature order corresponds to the rows of the associated matrix.mtx.gz file.
raw_feature_bc_matrix/matrix.mtx.gz
Gzip-compressed sparse UMI count matrix containing both tissue-associated and non-tissue-associated spatial barcodes.
Rows represent features and columns represent spatial barcodes.
HDF5 count matrices
anon/[SAMPLE]/filtered_feature_bc_matrix.h5
Filtered feature-by-barcode matrix stored in HDF5 format.
This file contains the same principal tissue-associated count data as the filtered_feature_bc_matrix/ directory but packages the matrix and annotations into a single file suitable for import by software such as Seurat, Scanpy and other 10x-compatible tools.
anon/[SAMPLE]/raw_feature_bc_matrix.h5
Raw feature-by-barcode matrix stored in HDF5 format.
This is the single-file equivalent of the raw_feature_bc_matrix/ directory and includes barcodes associated with both tissue-covered and off-tissue spots.
Spatial image and coordinate information
anon/[SAMPLE]/spatial/
Contains the image-registration, tissue-detection and spatial-coordinate outputs required to relate each expression barcode to its physical position on the Visium capture area.
spatial/aligned_fiducials.jpg
Quality-control image showing the detected Visium fiducial markers aligned to the capture-area image.
This image can be inspected to verify that Space Ranger correctly identified the fiducial frame and registered the spatial array to the supplied tissue image.
spatial/detected_tissue_image.jpg
Quality-control image showing the spots classified by Space Ranger as being under or outside tissue.
This image records the automated tissue-detection result used to define the filtered feature-by-barcode matrix.
spatial/scalefactors_json.json
JSON file containing scale factors that relate full-resolution image coordinates to the supplied high- and low-resolution tissue images.
It also contains scaling information for displaying the Visium spots and fiducial markers at the appropriate size.
spatial/tissue_hires_image.png
Downsampled high-resolution version of the tissue image used for spatial visualisation.
Spot coordinates can be mapped onto this image using the values in tissue_positions_list.csv together with the appropriate scale factor from scalefactors_json.json.
spatial/tissue_lowres_image.png
More highly downsampled version of the tissue image, intended for rapid plotting and interactive visualisation.
The image retains the aspect ratio of the original tissue image but has lower spatial resolution than tissue_hires_image.png.
spatial/tissue_positions_list.csv
Spatial position table linking each Visium spot barcode to its array and image coordinates.
For each barcode, the file records whether the spot was classified as under tissue, its row and column on the Visium array and its pixel row and column in the original full-resolution tissue image. This older Space Ranger filename generally contains no header row.
Sample-level summary and reference files
anon/[SAMPLE]/metrics_summary.csv
Comma-separated summary of the principal sequencing, mapping, tissue-detection and expression quality-control metrics reported by Space Ranger.
Metrics may include the number of spots under tissue, reads per spot, sequencing saturation, valid barcode and UMI fractions, median UMIs per spot and the number of detected features.
anon/[SAMPLE]/molecule_info.h5
HDF5 file containing molecule-level information for reads with valid spatial barcodes and UMIs that were assigned confidently to a feature.
The file records information underlying the count matrices, including the relationship between barcodes, UMIs, features and supporting reads. It can also be used as an input to certain downstream 10x Genomics pipelines.
This is a detailed molecule-record file rather than a sample-annotation table. Sample identity is represented by the enclosing directory name, RBJA_ST01 or RBJA_ST02.
anon/[SAMPLE]/probe_set.csv
Copy of the probe-set reference used when processing the sample.
The file describes the oligonucleotide probes and their associated target genes for the probe-based Visium assay. It provides the reference required to interpret how probe-level observations were combined into gene-level feature counts.
anon/[SAMPLE]/spatial_enrichment.csv
Feature-level spatial-autocorrelation results calculated by Space Ranger.
The table reports Moran's I and associated significance statistics for each eligible feature. Positive Moran's I values indicate that similar expression values tend to occur at neighbouring spatial locations, whereas values close to zero indicate little detectable spatial organisation.
Spatial enrichment is independent of the graph-based and K-means cluster assignments and can therefore identify spatially patterned features without first defining tissue clusters.
In standard Space Ranger output structures this file may be placed within the spatial/ directory. In this submitted archive it is located directly within the sample directory.
Notes on interpretation
- The
filtered_feature_bc_matrixfiles contain only spots identified as being under tissue and are generally the most appropriate starting point for downstream expression analysis. - The
raw_feature_bc_matrixfiles additionally contain off-tissue spots and are mainly useful for reassessing tissue calls, evaluating background signal or reproducing filtering procedures. - Graph-based and K-means clusters are automated, expression-based groupings. They should not be interpreted directly as cell types or anatomical regions without additional marker-based or histological assessment.
- t-SNE and UMAP coordinates describe similarity in expression space. Physical tissue locations are provided separately in
spatial/tissue_positions_list.csv. - Files under
RBJA_ST01andRBJA_ST02were generated independently. Cluster numbers, PCA coordinates, t-SNE coordinates and UMAP coordinates are therefore sample-specific and are not directly aligned between the two samples.
Code/software
N/A
Access information
Data was derived from the following sources:
- MCRC Biobank
Ethics approval and consent to participate
Human prostate tissue samples were obtained from the Manchester Cancer Research Centre (MCRC) Biobank under project 19 ROBR 01, approved by the MCRC Biobank Research Tissue Bank Ethics Committee (reference: 18/NW/0092). Glioblastoma tissue samples were obtained from the Salford Royal Hospital Biobank under ethical approval via the Integrated Research Application System (IRAS ID: 244538). All samples were collected with appropriate informed consent and handled in accordance with relevant ethical guidelines and regulations.
Human subjects data
All transcriptomic data are tertiary biological data (only counts tables and associated histology images). All associated Biobank IDs have been anonymised and all identifiable genomic information removed.
Human prostate tissue samples were obtained from the Manchester Cancer Research Centre (MCRC) Biobank under project 19 ROBR 01, approved by the MCRC Biobank Research Tissue Bank Ethics Committee (reference: 18/NW/0092). Glioblastoma tissue samples were obtained from the Salford Royal Hospital Biobank under ethical approval via the Integrated Research Application System (IRAS ID: 244538). All samples were collected with appropriate informed consent and handled in accordance with relevant ethical guidelines and regulations.