The analysis_summary.html file generated by the Xenium Onboard Analysis pipeline should be among the first output files a researcher looks at after running the Xenium instrument. This file serves three purposes:

1. Provides a convenient portable overview of data
2. Highlights summary metrics and plots that researchers can use to quickly QC data
3. Contains information that 10x Genomics support can use for troubleshooting

Where can I find the analysis summary?

The analysis_summary.html file can be found on the instrument after the run completes (see the Xenium Analyzer User Guide).

• After the run is complete, data generated across all the runs can be accessed under Menu > Open Settings > Runs.
• Click Open Output Folder Location to access the top-level output folder on the desktop. Click the individual runs to open a run-specific screen.
• To access the region-specific output folder, click Open Region Folder.
• The analysis_summary.html is available in the View Analysis Summary folder.

The analysis_summary.html file can be viewed on the instrument, on any web browser, and in Xenium Explorer.

At the top of the analysis_summary.html, there is a banner with the Run name and Region name, which were inputs on the instrument during the Initialize Instrument step (see the Xenium Analyzer User Guide). The Run start time, which is time-stamped by the instrument, is also shown (in UTC). There are five tabs: Summary, Decoding, Cell Segmentation, Analysis, and Image QC.

The default view begins on the Summary tab. When certain metrics fall below key thresholds, warning or error alerts will be shown below these tabs.

The Key Metrics panel is the starting point to QC your data and to spot outliers among multiple runs at a glance. These metrics are often helpful indicators if there is a problem with the data.

There are no universal thresholds for these metrics as interpretation requires some understanding of the sample and gene panel used. Researchers should have some understanding of how many cells are expected given the tissue size and type. Here are some important considerations for interpreting key metrics:

• Median transcripts per cell and decoded transcript density will depend on tissue type and the selected gene panel (e.g., a 300-gene panel will have more transcripts than a 50-gene panel).
• The number of cells detected will depend on tissue type and the size of the region of interest (ROI) selected on the instrument in the analysis.

The Sample Region Summary (left) shows the Region name, Slide ID, and Cassette name, which are required inputs during instrument initialization (see the Xenium Analyzer User Guide). Preparation method, also input during instrument initialization, is optional.

Every analysis_summary.html file has an overview scan (right), which shows the nuclei-stained (DAPI) image. The slide ID is displayed below the overview scan image (ID:). This ID location indicates the slide orientation, as it matches the relative location of the ID on the physical Xenium slide.

The purpose of the image is to help researchers quickly identify the tissue region on the slide that was used to generate the data in the Xenium output bundle. For example, if you have three mouse brain replicates on a slide, the overview scan quickly shows which among the top, middle, or bottom replicates the summary metrics correspond to.

The Region Details view shows a DAPI image of the analyzed region along with a colored overlay of counts of high-quality decoded transcripts per bin. The overlay can be adjusted to various bin sizes and opacity scales as needed. Finer spatial resolution of decoded transcript density is shown with smaller bin sizes. If moused over, a zoomed-in DAPI image will be shown.

This view can be used to QC a variety of common sample preparation issues:

• Does the DAPI morphology image look as expected?
• Were there any tears in the tissue or detachment of tissues from the slide?

There are other questions that can be addressed using this view that require further experimental context, such as knowledge of the sample type. For example, does the general transcript density match up with the tissue morphology? If the tissue is a mouse brain, one would expect to see particularly high density in the hippocampus.

The Run Information section is meant to help researchers quickly determine key metadata regarding the run.

The Software section shows key metadata for troubleshooting purposes. 10x Genomics continually updates the instrument and analysis software on the Xenium instrument to introduce new features, fix bugs, and provide the best user experience.

The Panel Specification section displays the Xenium panel that was used for the experiment. If the QC metrics look drastically different from what is expected, it may be prudent to double-check that the correct panel was used. The source of the metadata is the gene_panel.json file. If running a pre-designed panel, the JSON is already on the instrument. If running a custom or add-on panel, the JSON file would be produced by the Xenium Panel Designer or 10x support team and loaded onto the instrument during instrument initialization using a USB drive (see the Xenium Analyzer User Guide).

If the data were analyzed with Xenium Ranger, a new panel will display information about the analysis:

• Run ID
• Run timestamp
• Xenium Ranger software version
• The command(s) used to analyze the data

The Xenium codebook contains a collection of codewords that are assigned to genes in a gene panel. The pipeline uses the gene_panel.json to specify a given gene name to an indexed codeword. Each codeword is defined based on a pattern of fluorescent signal intensities recorded across channels and cycles. Some codewords are reserved for negative controls. For more information about transcript decoding, see the decoding, quality scores, and controls section of the Xenium algorithms page.

The Decoding Yield section provides QC metrics at a glance. Biological and experimental context may be needed for interpretation.

Below are descriptions for the Negative Controls metrics. The adjusted negative control probe rate measures both erroneous decoding and off-target binding, so it will always be greater than or equal to the adjusted negative control codeword rate. Both depend on the total number of high-quality transcripts detected, so a high value may result from lower detection of gene transcripts rather than nonspecific binding. For more information about negative controls, see the decoding, quality scores, and controls section of the Xenium algorithms page.

The "Negative control probe counts per control per cell" metric is derived first from calculating the absolute number of negative control probe counts and then calculating the metric per control probe per cell. The "Estimated number of false positive transcripts per cell" is derived from this metric. Both metrics are available in outputs generated with Xenium Onboard Analysis v1.6 and later. Derived values may differ slightly from those reported in summary outputs due to using rounded values. They are calculated using the following inputs:

• Total high quality decoding transcripts: displayed on the Summary and Decoding tabs, includes genes (Q-Score ≥ 20) and negative controls
• Percent of transcripts within cells: displayed on the Cell Segmentation tab
• Adjusted negative control probe rate: displayed on the Decoding tab
• Number of target genes: the sum of genes in the pre-designed panel and any add-on custom genes, displayed on the Summary tab
• Number of negative control probes in the panel: each pre-designed and add-on panel contains 20 negative control targets (except runs that use the Xenium Mouse Brain base panel, which have 27 negative control targets).
• Number of cells detected: hero metric on the Summary tab

1. The Negative control probe count is calculated as:

$\footnotesize{\text{Q20 decoded transcripts in cells}\times{\text{Adjusted negative control probe rate}}\times{\frac{\text{Number of negative control probes}}{\text{Number of target genes}}}}$

where Q20 decoded transcripts in cells is calculated as:

$\footnotesize{\text{Total high quality decoding transcripts}\times{\text{Percent of transcripts within cells}}}$

2. The Negative control probe count per control per cell is then calculated as:

$\footnotesize{{\text{Negative control probe count}}\div{\text{Number of negative control probes}}\div{\text{Number of cells detected}}}$

3. The Estimated number of false positive transcripts per cell is calculated as:

$\footnotesize{{\text{Negative control probe counts per control per cell}}\times{\text{Number of target genes}}}$

The Counts per Gene plot shows the total number of transcripts passing quality thresholds for every gene in the panel. This plot is a histogram of absolute counts of decoded transcripts (x-axis) by gene (y-axis). Genes are listed in alphabetical order. Hover over each bar to see which gene it corresponds to. This plot can be used to quickly compare multiple samples analyzed with the same panel, or to quickly compare add-on vs. pre-designed panels.

The Transcript Quality plot shows the distribution of quality scores for decoded transcripts, separating out transcripts that decode to panel genes, negative control probes, negative control codewords, and unassigned codewords. The transcript counts (y-axis) can be viewed in linear or log scale. High-quality samples should have panel genes with high quality scores (≥ 20) and negative controls of both types with low quality scores.

Click on a gene or negative control in the legend to hide it from view. This plot can be used to ascertain how much data is lost due to quality filtering. Data with quality scores < 20 are filtered from the cell-feature matrix, and downstream analyses using the matrix file, but these filtered transcripts can still be found in the transcripts files. For more information, see Understanding Xenium Outputs.

For each gene, the Gene-Specific Transcript Quality plot shows the total number of decoded transcripts of any quality (x-axis) against the mean quality of those decoded transcripts (y-axis). The x-axis (total transcripts per gene) is plotted on a log10 scale, while the y-axis is on a linear scale.

When interpreting this plot, one would typically expect to see all or most genes have mean quality scores ≥ 20, and the controls should be < 20; in other words, genes should be at the top-right quadrant and controls should be in the bottom-left quadrant. Examine genes, especially custom genes, in the panel for any performance problems (i.e., low density or low mean Q-Score may indicate problems in the performance of a given gene).

Negative controls for both decoding and probes should have low transcript counts and low quality:

• A high rate of negative control codewords may indicate possible imaging issues, such as autofluorescence in one or more cycle-channels that are part of a negative control codeword. Check the RNA images in the aux_outputs/ directory.
• High quality and count of all negative control probes may indicate assay workflow issues, such as nonspecific probe hybridization or ligation conditions. It is ok if only a few negative control probes show high quality and/or count.

The Mean Q-score Spatial Map plot shows binned transcript quality scores across the sample area. This plot can be used to identify any technical artifacts associated with one or more fields of view (FOVs). The example screenshot below is from a pancreatic cancer dataset:

The purpose of cell segmentation is to approximate cell boundaries so that transcripts can be assigned to cells. Downstream, these results will be used to produce a cell-feature matrix, similar to those output by existing single cell and spatial technologies. For more information, see the cell segmentation section of the Xenium algorithms page.

Here are descriptions for the Segmentation Metrics:

For all cells with transcripts, the Cell Size Distribution view shows a histogram of cell area in µm². The area is computed from the cell segmentation mask.

The Genes per cell view shows a histogram of the total number of unique, panel (non-control) genes found in each cell for all cells with transcripts. The Transcripts per cell view shows a histogram of the total number of transcripts found in each cell over all panel (non-control) genes for all cells with transcripts.

The Transcripts Per Cell view shows the spatial distribution (left) and UMAP projection (right) of cells colored by the total number of transcripts detected in each cell. For very large samples, a subset of cells may be plotted for better visualization performance.

The Clustering view shows the spatial distribution (left) and UMAP projection (right) of cells colored by cluster assignment using the Onboard Analysis pipeline's automated clustering algorithm. The clusters should reflect groups of cells that have similar expression profiles. In the left plot, cells are colored according to their cluster assignment and plotted in their spatial location. Only cells with a nucleus detected by the DAPI stain are used in the clustering algorithm.

In the right plot, the axes correspond to the 2-dimensional embedding produced by the UMAP algorithm. In this space, pairs of cells that are close to each other have more similar gene expression profiles than cells that are distant from each other.

The Top Features by Cluster table displays results from the Onboard Analysis pipeline's automated differential expression analysis. For each cluster, the table shows features that are more highly expressed in that cluster relative to the rest of the sample.

A differential expression test was performed between each cluster and the rest of the sample for each feature.

• The Log2 fold-change (L2FC) is an estimate of the log2 ratio of expression in a cluster to that in all other cells. A value of 1.0 indicates a 2-fold greater expression in the cluster of interest.
• The p-value is a measure of the statistical significance of the expression difference and is based on a negative binomial test. The p-value reported here has been adjusted for multiple testing via the Benjamini-Hochberg procedure.

In this table, you can click on a column to sort by L2FC or p-value for each cluster. Features were filtered (mean object counts > 1.0) and the top N features by L2FC were retained for each cluster. Features with L2FC < 0 or adjusted p-value ≥ 0.10 are grayed out. The number of top features shown per cluster, N, is set to limit the number of table entries shown to 10,000 (N=10,000/K² where K is the number of clusters). N can range from 1 to 50. For the full table, please refer to the differential expression CSV files produced by the pipeline.

The gallery in this tab shows a stitched and reconstructed fluorescence image of the analyzed region obtained by downsampling and processing the internal sensor data for each cycle (row) and channel (column).

The intensity values in the thumbnail image give the average fluorescence intensity measured by the sensor in units of photoelectrons (pe) per square micron of tissue area. They are downsampled to 200 x 200 pixels from the high-resolution images available in aux_outputs/per_cycle_channel_images/.

This gallery can be useful for quick data QC, such as checking for stitching errors, total cycle dropouts or oversaturation, debris, or low transcript density areas. Patterned changes in fluorescence intensities across cycle-channel plots are a signal of good image quality. For example, in the images below, certain tissue areas are dim in some cycles and bright in others:

Fluorescence signals that are always high in the same location in all cycles indicate tissue autofluorescence and may affect transcript data at that location.