Use the Xenium Pipeline

	Abbreviations Key
AnnData	annotated data [Python/R package for storing spatial matrices]			MEX	market exchange [format]
CSV	comma-separated values			MIP	maximum-intensity projection
FFPE	formalin-fixed paraffin-embedded [tissue preservation method]			OME	open microscopy environment
gz	gzipped			PCA	principal component analysis
H5AD	.h5 AnnData [format]			QC	principal component analysis
HDF5	Hierarchical Data Format, version 5 [proprietary file format]			scRNA-seq	single-cell RNA sequencing
H&E/IF	hematoxylin and eosin immunofluorescence [images]			STalign	spatial transcriptomics alignment
HISE	Human Immune System Explorer			UMAP	uniform manifold approximation and projection
IDE	integrated development environment			VM	virtual machine
ISD	instrument sensor data

At a Glance

The Xenium pipeline enables spatial transcriptomics analysis within HISE, converting raw Xenium data into formats that can be analyzed, explored, and visualized in HISE.

This document describes the pipeline in three broad stages or steps, summarized in Table 1 and depicted in the following image.

 Preprocessing

It's beyond the scope of this document to cover all of the preprocessing functions in detail, but let's briefly explore how Xenium handles data during this stage. For each FFPE tissue sample, Xenium generates two forms of raw data:

1. Initial raw data. First, the 10x machine generates a massive amount of raw ISD data. One Xenium slide with the entire imageable area selected produces its own directory ranging from 7–60 GB, depending on the tissue and the panel. A single run typically contains four slides, for a total of 28-240 GB. (For examples of public datasets, see the 10x Genomics website.) The raw Xenium output files are listed in Table 2.

2. Preprocessed raw data. The machine then parses the data in a smaller raw data set that contains decoded transcript information. This transient data remains active only until cell segmentation, at which point it's also archived. A directory containing the resulting machine-processed data is created. The Xenium preprocessing output directory includes the files listed in Table 3.

TABLE 3
Stage	Output	Example
Preprocessing	xenium_<tissue>_adata_filtered.h5ad	Not pictured
QC	PDF reports (for example, nucleus/cell area plots)
Cell labeling	Cell-type predictions
Neighborhood analysis	Spatial cluster plots created using CellCharter
Summary report	<tissue>_pipeline_summary.html

Ingestion

This is where HISE enters the picture. The necessary configuration files are ingested (Table 4). Then the preprocessed Xenium files (see Table 3) are moved into a watchfolder, which triggers ingestion of the data into HISE. (Be sure to follow this sequence, since the pipeline run will fail if the tar file is ingested before the configuration files.)

The data ingestion workflow, including preprocessing, is shown in the accompanying figure.

During ingest, the directory is unzipped into /processed_data/. The files in this new directory are listed in Table 5. 

Unlike other types of data, Xenium data doesn't require a sample or submission sheet. You can simply ingest the raw data into HISE, which handles organization, validation, and metadata extraction for you. A filename looks something like this:

202208311221_EXP-00422-LN-FFPE-NDGFKF_XETG00123_region_A1

Output of results

Table 6 contains a list of downloadable/servable result file types. 

Visualization of results

You can use an interactive visualization tool to understand your results. Examples of the types of visualizations you can create are listed in Table 7.

 Exploration

After ingestion, your data is ready for exploration and interactive analysis in a HISE NextGen IDE.

Interactive data analysis

The Jupyter Notebook/IDE environment is used for intersample analysis in either interactive mode (manually interacting with a Jupyter notebook) or batch mode (notebook jobs). You can load AnnData (.h5ad) files and other outputs for custom analysis using Python libraries such as Scanpy, Squidpy, or Seaborn. You can also perform dimensionality reduction (for example, UMAP, t-SNE, or PCA) and clustering. Another option is to run an advanced query to filter cells by type, spatial region, or gene expression.

For deeper biological insights, you can combine Xenium data with other spatial transcriptomics datasets, such as Visium, or with scRNA-seq data. For cross-dataset analysis, batch correction, or spatial alignment, you can integrate tools like Scanpy, Squidpy, or STalign. Then export your results for further downstream analysis or publication.

Related Resources

Submit and Monitor Pipeline Batches (Tutorial)

Understand Automated Pipelines

Configure a Pipeline (Tutorial)