I'm wondering if different versions of cell-ranger (say v 2.0 and v 7.0) can result in substantially different cell-barcodes, or UMI counts. I ask this because I've tried to reproduce results from a paper that used v 2.0, but the results (cell and UMI counts) are very inconsistent.
Yes, absolutely they can! 10x has changed the code and parameters substantially from cellranger 2 to 7. You can read their release notes to see the descriptions of what has changed, but since cellranger 3, it has been a closed source project, so you cannot really know. Unless you have a specific reason to use cellranger, I’d suggest using another tool such as alevin-fry (disclaimer; I’m an author on that paper), or STARSolo, whose operation is well documented and whose code is open source.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
Thanks for your reply, Rob. I did read the release notes and found that there's been substantial changes, but was surprised to see that even v3 and v2 output very different matrices 🤔
Yes; in addition to the big changes (e.g. version 7 counts intronic reads by default, while prior versions did not), there are certainly many smaller changes that, nonetheless, can have significant effect on the resulting counts. Yet, we do not know the actual / full nature of those changes since the tool itself is now closed-source.
Wow! I spent an entire week worrying about not reproducing results of a paper that used version 2.0, since I used version 7 and obtained vastly different DEGs (confirmed this wasn't a problem with my methods since I obtained the same result after using the reported count matrix). Just wondering what this implies for earlier published works that used version 2.0 e.t.c and how this could have impacted the results/interpretations.
I will definitely try out the tools you suggested to see which ones return the most consistent results.