SCbrowse is an interactive browser for single-cell ATAC-seq data. It allows to simultaneously explore the accessibility patterns of selected cells in the embedding space and for selected genomic regions. The tools freely available under a GNU Lesser General Public License v3 or later (LGPLv3+).
A live example showing Zebrafish embryo scATAC-seq data is available at scbrowse.mdc-berlin.de with data from McGarvey, Kopp, Vučićević, et al. biorxiv 2020.
It is possible to select cells manually by selecting the "Lasso select" or "Box select" in the navigation bar above the embedding scatterplot. As a result, the area underneigth the selection will appear highlighted.
Yes.
Above the scatter plot visualization, there are two buttons that allow to undo the last selection or all selections.
On the top right, you can search and select a gene of interest or alternatively select genomic coordinates directly. Using the zoom in/out and move functionality, allows to navigate and adjust the view along the locus.
The following lines show how to install scbrowse and its requirements using conda
conda create -n scbrowse python=3.7 conda activate scbrowse conda install -c bioconda bedtools git clone https://github.com/BIMSBbioinfo/scbrowse pip install -e scbrowse[gunicorn]
SCbrowse needs two required ingredients:
- A dataset stored in h5ad format which includes a raw countmatrix, cell annotation and precomputed embeddings.
- Gene annotation in bed format
To prepare the dataset, set up an AnnData dataset adata = anndata.AnnData(...): 1. adata.X should contain a regions by cells raw sparse count matrix 2. adata.obs should contain the fields chrom, start and end, which define the genomic bins. 3. adata.var should contain the zero, one or several cell annotation tracks (categorical). 4. adata.varm should contain a 2D cell embedding stored as numpy array. 5. adata.uns should contain a entry 'embeddings' which lists the available keys for storing the embeddings in varm.
After preparation, the AnnData should look like this:
>>> from anndata import read_h5ad
>>> adata = read_h5ad('scbrowse.h5ad')
>>> adata
AnnData object with n_obs × n_vars = 1345115 × 23008
obs: 'chrom', 'start', 'end'
var: 'sample', 'cell_label', 'strain'
uns: 'embeddings'
varm: 'UMAP_all_cells'
>>> adata.uns['embeddings']
array(['UMAP_all_cells'], dtype=object)
Furthermore, to compile a countmatrix, consider using scregseg. It offers functions for creating a countmatrix, including scregseg make_tile, scregseg bam_to_counts or scregseg fragments_to_counts as in the example below. For example for a 1000 kp resolution countmatrix use:
scregseg make_tile --bamfile <bam> --resolution 1000 --bedfile <outputbed> scregseg bam_to_counts --bamfile <bam> --counts <countmatrix> --regions <outputbed>
The best way to obtain cell embeddings is by first apply dimensionality reduction, e.g. by running cisTopic, SnapATAC or any other tool of your choice. Then create a 2D UMAP or t-SNE embedding for visualization and save it in the AnnData slot.
To launch scbrowse in development mode, first, define the following environment variables to point to the input files
export SCBROWSE_MATRIX=<matrix.h5ad> export SCBROWSE_GENES=<genes.bed> export SCBROWSE_LOGS=<scbrowse.log> scbrowse
Afterwards you can browse the data locally in a web-browser by opening https://localhost:8051
The simplest way to deploy scbrowse is via gunicorn.
The repository contains a startup.sh script
that launches the web server.
It needs to be adjusted to point to the input file locations.
Then run
sh startup.sh
and open the scbrowse in a web browser at https://localhost:8000