Skip to main content
Premium Trial:

Request an Annual Quote

10x Genomics Study Highlights Xenium Data With Cross-Platform Validation

Premium

NEW YORK – A new spatial transcriptomics dataset generated by 10x Genomics with its forthcoming Xenium platform demonstrates its concordance with existing 10x single-cell and spatial technologies.

Included as part of a preprint posted to BioRxiv this month, the downloadable data, from a 313-gene human panel, could help convince potential customers that they'll get good results with Xenium, according to Raphael Gottardo, a bioinformatician at the Swiss Institute of Bioinformatics.

"What I love about these preprints is the availability of data, so we can look at it ourselves," he said, noting that the data appeared to be "high quality," based on parallel results from 10x's single-cell sequencing method and from Visium, its other spatial transcriptomics platform. "Seeing multiple modalities agree and give you similar results makes you more confident," he said. He was also impressed that Xenium allowed for other assays, such as immunofluorescence or hematoxylin and eosin (H&E) staining.

"You can probably even do Visium afterwards," he said, noting that he had not confirmed this with 10x.

The data should help 10x make up ground in the chase for customers in the high-resolution spatial genomics market. But the study also raises questions about cell segmentation methodologies and how they may affect this competition.

In the study, 10x researchers applied Xenium, Visium, and single-cell sequencing technologies to analyze the breast cancer tumor microenvironment. It employed a 280-gene human breast panel that will be available at launch as well as custom gene targets. The study also provided data from 10x's new single-cell gene expression assay for formalin-fixed paraffin-embedded samples.

"In this particular case, we're able to see some additional molecular detail that's maybe hidden in an H&E stain," said Benjamin Hindson, cofounder and CSO at 10x. 

The authors reported data on 167,885 cells, approximately 37 million total transcripts, and a median of 166 transcripts and 34 genes per cell.

In order to recreate the gene-cell and gene-spot matrices used in data analysis for its other platforms, 10x researchers had to define cell boundaries and assign transcripts to cells. They did so using DAPI, a staining molecule that permeates the nuclear membrane and binds DNA, and "expanded outwards until either 15 μm maximum distance was reached, or the boundary of another cell was reached," according to the study.

With the caveat that he normally works with already segmented data, Gottardo said this approach was "a fairly simple analysis, from my point of view."

An alternative would be a multimodal approach, such as the one NanoString has adapted for its CosMx high-resolution spatial omics platform, using not only DAPI but also cell membrane markers as well as transcript-based boundary refinement. The Baysor algorithm, developed for use on multiplex fluorescence in situ hybridization-based spatial methods, is another option. Both 10x and NanoString purport to use advanced machine learning algorithms as part of the segmentation process.

The segmentation method described in the study will be the standard offering at launch, a 10x spokesperson said, adding that its deep learning algorithm is trained on images from specific tissue types relevant to the Xenium panels, as well as additional tissue types. "For the future, we are investing significantly in our segmentation roadmap in order to continue to build out our DAPI-based models, as well as full-cell segmentation based on other stains," the spokesperson said.

"Whoever brings the best tools to go with the platform will have a big advantage," Gottardo predicted.

He also predicted that the study will help 10x catch up to NanoString Technologies and Vizgen in the race to provide spatial genomics technologies with single-cell resolution.

"They were a little bit behind in releasing this product," he said. "But releasing multiple assets and datasets at the same time, they're catching up pretty fast. Either they're now in front, or not far behind."