Prepare Data

Data Format Requirements

If your data is an h5ad file and meets the following requirements, you can go straight to cellxgene launch.

• Matrix data (usually raw or normalized expression values) in anndata.X
• At least one embedding (e.g., tSNE, UMAP) in anndata.obsm, specified with the prefix X_ (e.g., by default scanpy stores UMAP coordinates in anndata.obsm['X_umap'])
• A unique identifier is required for each cell, which by default will be pulled from the obs DataFrame index. If the index is not unique or does not contain the cell ID, an alternative column can be specified with --obs-names
• A unique identifier is required for each gene, which by default will be pulled from the var DataFrame index. If the index is not unique or does not contain the gene ID, an alternative column can be specified with --var-names

What about R objects from Seurat or Bioconductor!? You can use sceasy to convert to h5ad. Seurat also has conversion tools that you can try.

Can I use data hosted on the web somewhere? Yes! You can launch from a URL instead of a filepath. The same data format requirements apply. Please see here

Data Format Options

Category Colors

cellxgene will display scanpy-style color information for category-label pairs. An example of this format is shown below:

>>> category = "louvain"
>>> # colors stored in adata.uns must be matplotlib-compatible color information
>>> adata.uns[f"{category}_colors"]
array(['#1f77b4', '#ff7f0e', '#2ca02c', '#d62728', '#9467bd', '#8c564b', '#e377c2', '#bcbd22'], dtype='>> # there must be a matching category in adata.obs
>>> category in adata.obs
True

To test that you've done this properly, check that for your given category the number of colors match the number of category values and that the second command below results in a mapping from categories to colors.

>>> len(adata.obs[category].cat.categories) == len(adata.uns[f"{category}_colors"])
True
>>> dict(zip(adata.obs[category].cat.categories, adata.uns[f"{category}_colors"]))
{'CD4 T cells': '#1f77b4', 'CD14+ Monocytes': '#ff7f0e', 'B cells': '#2ca02c', 'CD8 T cells': '#d62728', 'NK cells': '#9467bd', 'FCGR3A+ Monocytes': '#8c564b', 'Dendritic cells': '#e377c2', 'Megakaryocytes': '#bcbd22'}

You can disable this feature using the --disable-custom-colors flag for cellxgene launch. cellxgene will then chose colors from its standard color palettes.

Using cellxgene prepare

If you need to normalize your data and create an embedding, cellxgene can do that for you with the prepare command.

What is cellxgene prepare?

cellxgene prepare offers an easy command line interface (CLI) to preliminarily wrangle your data into the required format for previewing it with CELLxGENE Annotate. It runs scanpy under the hood and can read in any format that is currently supported by scanpy, including mtx, loom, and more listed in the scanpy documentation.

prepare uses scanpy to:

• Handle simple data normalization (from a recipe)
• Do basic preprocessing, run PCA, and compute the neighbor graph
• Generate UMAP and tSNE embeddings
• Infer clusters using Louvain

You can control which steps to run and their methods (when applicable), via the CLI. The CLI also includes options for computing QC metrics, enforcing matrix sparcity, specifying index names, and plotting output.

What is cellxgene prepare not?

cellxgene prepare is not meant as a way to formally process or analyze your data. It's simply a utility for quickly wrangling your data into cellxgene-compatible format and computing a "vanilla" embedding so you can try out cellxgene and get a general sense of a dataset.

Quickstart for cellxgene prepare

pip3 install cellxgene[prepare]
cellxgene prepare dataset.h5ad --output=dataset-processed.h5ad

This will load the input data, perform PCA and nearest neighbor calculations, compute UMAP and tSNE embeddings and louvain cluster assignments, and save the results in a new file called dataset-processed.h5ad that can be loaded using cellxgene launch.

Example usage

As a quick example, let's construct a command to use prepare to take a raw expression matrix and generate a processed h5ad ready to visualize with CELLxGENE Annotate.

We'll start off using the raw data from the pbmc3k dataset. This dataset is described here, and is available as part of the scanpy package. For this example, we'll assume this raw data is stored in a file called pbmc3k-raw.h5ad.

Our prepare command looks like this:

cellxgene prepare pbmc3k-raw.h5ad \
    --run-qc \                                  # (A)
    --recipe seurat \                           # (B)
    --layout tsne --layout umap \               # (C)
    --output pbmc3k-prepared.h5ad               # (D)

Let's look at what prepare is doing to our data, and how each step relates to the command above. You can see a walkthrough of what's going on under the hood for this example in this notebook.

• (A) - Compute quality control metrics and store this in our AnnData object for later inspection
• (B) - Normalize the expression matrix using a basic preprocessing recipe (auto) - Do some preprocessing to run PCA and compute the neighbor graph (auto) - Infer clusters with the Louvain algorithm and store these labels to visualize later
• (C) - Compute and store UMAP and tSNE embeddings
• (D) - Write results to file_

Options for cellxgene prepare

For the most up-to-date and comprehensive list of options, run cellxgene prepare --help --embedding controls which dimensionality reduction algorithm is applies to your data. Options are umap and/or tsne. Defaults to both.

--recipe controls which normalization steps to apply to your data, based on one of the preprocessing recipes included with scanpy. These recipes include steps like cell filtering and gene selection; see the scanpy documentation THIS LINK IS BROKEN for more details. Options are none, seurat, or zheng17. Defaults to none.

--sparse is a flag determines whether to enforce a sparse matrix. For large datasets, prepare can take a long time to run (a few minutes for datasets with 10-100k cells, up to an hour or more for datasets with >100k cells). If you want prepare to run faster we recommend using the sparse option. If this flag is not included, default is False

--skip-qc by default, cellxgene prepare will compute quality control metrics (saved to anndata.obs and anndata.var) as described in thescanpy documentation THIS LINK IS BROKEN. Pass this flag if you would like to skip this step.

--make-obs-names-unique / --make-var-names-unique determine whether to rename obs (cell) / var (gene) names, respectively, to be unique. Default is True.

--set-obs-names controls which field in anndata.obs (cell metadata) is used as the index for cells (e.g., a cell ID column). Default is anndata.obs.names

--set-var-names controls which field in anndata.var (gene metadata) is used as the index for genes. Default is anndata.var.names

--output and --overwrite control where the processed data is saved.