Compatibility of scRNA-seq data from different GRCh38 versions (2020-A vs 2024-A)

[Captain-Pam]

Hi,

Thank you for your tools.

I have scRNA-seq data processed with cellranger-count GRCh38-2024-A (primary data) and GRCh38-2020-A (13 public samples without fastq files). From the 10x reference release notes, I understand these versions differ not just in gene counts but also in gene annotations (e.g., renamed/added/removed genes, transcript updates).

Can these datasets be directly merged/integrated (e.g., via merge() or IntegrateData() in Seurat) without other processing, given the annotation differences? Taking the intersection would probably cause some genes in my most sample data to be lost. Is it necessary to remove this small part of the sample? or is there a robust way to harmonize them?

Thanks for your insights!

[samuel-marsh]

Hi,

As this isn't bug/issue with Seurat package per say I'm going to move this post to discussions so that others in community can weigh in as well but will add my personal thoughts as well below.

Best,
Sam

If the data is not available to be reprocessed then yes things can be a bit tricky. My personal recommendation would be to update the gene symbols in both sets of data to the most recent (to ensure you can match as many as possible) and then examine things bit further.

To update gene symbols to latest you can use helper function in from my package scCustomize (see vignette here: https://samuel-marsh.github.io/scCustomize/articles/Update_Gene_Symbols.html). After doing that I would create Seurat objects (so that non-expressed or lowly expressed genes are filtered out by min.cells parameter) and then look to see how many genes intersect and how many are lost from each alignment. Then you can decide how to proceed.

I would also email authors just asking for access to fastq from public datasets because you are right and having everything aligned with the same annotation and then same version of cell ranger will take away a batch variable that can be tricky to navigate.