FindMarkers reporting different genes in v5 / backwards compatibility of seurat object?

[mahogny]

Hi!

We were recently forced to update to Seurat v5 from v4 or v3 -- unfortunately not clear which version, and this doesn't help the troubleshooting

Now FindAllMarkers report very different genes from before, to the extent that I question if this is random or not (based on plotting genes on UMAPs). It's hard to go back and do a proper comparison unfortunately

Has anything been changed here that could affects things? We use default settings for this function. Might it be using the wrong data layer or something?

Notably: We are loading an old seurat RDS with the new version. I imagine tons can go wrong here. But much processed data published out there /is/ in the old format, so I think seurat will need a sensible route to update such old objects and be backwards compatible. Is this the plan? I think some general recommendation might be due in terms of how we should release processed data in the future (besides just raw counts and our code)

Best,
Johan

[yaolutian]

And also, V5 seems generate much higher fold change and more DEGs than V4. which setting is different?

[Shuresearcher]

Do you know why is it? Can we trust DEGs from V5?

[Yanst86]

They switched the algorithm for differential gene expression - https://satijalab.org/seurat/articles/announcements.html
In addition, please note that for FindAllMarkers they changed the min.pct threshold to 0.01 instead of 0.1 - that's why you also get genes that are expressed in a very low fraction of the cells that you didn't see at all in Seurat V4.

[mahogny]

Great answer, thanks! It probably would be best if the previous min.pct was kept for backwards compatibility!

[Shuresearcher]

Does this changed affect fold change of DEGs？

[sarahhp]

I noticed the algorithm change (documentated in help), but not the min.pct! Thanks

[sarahhp]

Also logfc.threshold changed from 0.25 to 0.1 with v5.