Single Cell Visualizations

Authors: Batuhan Cakir, Simon Murray and Vladimir Kiselev.

If you would like to make your single-cell RNA-seq data publicly available on a website, for example as a supplement for a publication, we can help you with that!


We use cellxgene to visualize single cell RNA-seq data. cellxgene is an interactive data explorer which is very scalable and flexible.

To be able for us to create a cellxgene website for your data we need to have your data in the h5ad (AnnData) format.


AnnData format usually contains the following slots:

  • X contains the expression matrix.
  • obsm contains the embeddings data.
  • obs contains the cell metadata.
  • var contains the gene metadata.

When you work with cellxgene you only need to modify two of the slots above: obsm and obs.

X slot

cellxgene works faster when the expression matrix is stored in CSC (compressed sparse column) format instead of CSR (compressed sparse row) format or dense Numpy array (which sometimes can create a smaller h5ad file depending on the sparsity of your data).

To convert your expression matrix into the CSC format please use:

adata.X = scipy.sparse.csc_matrix(adata.X)

To convert your expression matrix into the Numpy array please use:

adata.X = scipy.sparse.csc_matrix.toarray(adata.X)

obsm slot

To visualize your cells in 2D cellxgene uses obsm slot. If there are multiple embeddings stored in this slot they will all be available on the web interface.


cellxgene requires that all of the embeddings’ names are prefixed with X_. For example, X_umap, X_pca or X_some_embedding.

obs slot

To highlight and colour your cells cellxgene uses obs slot. The colouring will depend on the type of you cell metadata contained in the obs slot.

When the metadata is categorical, i.e. there is one colour per category, the visualization will look like this:


To make your cell metadata categorical please use the following code:

import pandas as pd
adata.obs['metadata_name'] = pd.Categorical(adata.obs['metadata_name'])

When the metadata is continuous, the visualization will look like this:



Note there is a continuous scale on the right side of the plot.

To make your cell metadata continuous please use the following code:

import numpy as np
adata.obs['metadata_name'] = np.float32(adata.obs['metadata_name'])

Data Conversion

We have released the sceasy package on GitHub ( to easily convert other single-cell file types to AnnData format for visualization with cellxgene. Currently it supports converting Seurat, SingleCellExperiment and Loom objects to AnnData. By default it transfers expression matrices, cell and gene metadata table, and, if available, cell embeddings in reduced dimensions to AnnData.


Before installing the conda packages below please first create a new conda environment EnvironmentName and activate it. Everything else can be installed in R.

sceasy is installable either as a bioconda package:

conda install -c bioconda r-sceasy

or as an R package:


To use sceasy ensure the anndata package (version has to be < 0.6.20) is installed:

conda install anndata == 0.6.19  -c bioconda

In addition, please also ensure the loompy package (loompy version < 3.0.0) is installed:

conda install loompy == 2.0.17  -c bioconda

You will also need to install reticulate package:


Finally, before converting your data please load the following libraries in your R session:

loompy <- reticulate::import('loompy')

Seurat to AnnData

sceasy:::convertFormat(seurat_object, from="seurat", to="anndata",

Seurat to SingleCellExperiment

sceasy:::convertFormat(seurat_object, from="seurat", to="sce",

SingleCellExperiment to AnnData

sceasy:::convertFormat(sce_object, from="sce", to="anndata",

SingleCellExperiment to Loom

sceasy:::convertFormat(sce_object, from="sce", to="loom",

Loom to AnnData

sceasy:::convertFormat('filename.loom', from="loom", to="anndata",

Loom to SingleCellExperiment

sceasy:::convertFormat('filename.loom', from="loom", to="sce",