New version of cellranger-atac breaks frag size distr calculation.

[joseverdezoto]

Hey Kai,

Thanks again for making this great package. I've been realigning some of my 10X snATAC data with the updated cellranger-atac v2.2.0. In this new version of cellranger-atac, the strand info is added as the last column of the fragment file. I just saw in the docs that if this is the case, fragment_single is added to the .obsm instead of fragment_paired.

This is my fragment file:

When I load it and try to compute the distribution of fragment sizes, I get the following error:

The error goes away if I use fragments from multiome data (no strand info and only 5 columns which does have fragment_paired in the .obsm) so I'm sure this is the issue.

What would be the correct way to approach this? I'd really appreciate your input as I'm certain I have paired-end data.

Best,
Jose

[kaizhang]

What does the new column (strand) mean for fragments? Does 10x have a documentation for this?

[joseverdezoto]

Hey Kai,

Thanks for the prompt response! This is what the release notes for cellranger-atac v2.2.0 (released ~1 month ago) say:

And this is the more detailed description of the .tsv fragment files fields:

If I'm correct, all 10x snATAC-seq data should be paired-end, so would a change to the way the fragments are imported be needed? How should I approach this? I realigned my data with the previous cellranger-atac in the meantime.

[kaizhang]

Yes, I will need to update the code to account for this change. Until a fix is available, a workaround for you is to simply drop the 6th column from the fragment file.

[agupta229]

Hi @joseverdezoto, thanks for opening up this thread! I recently ran into the same issue and was able to drop the 6th column to see fragment_paired be detected.

However, I'm still running into issues with calculating the fragment size distribution and seeing associated plots. While I was able to have this work in the single sample tutorial code, it does not work for me with the multisample tutorial, nor with my own data.

If I try to pass my full adatas object, I get TypeError: cannot pickle 'builtins.AnnData' object

If I try to use a for loop to generate a series of plots, one for each adata object, I get the following error:

I've also tried just passing adatas[0] as my data object, but that also fails.

Interestingly, my TSS enrichment plots and such are all working, so this seems specific to this fragment size distribution calculation. It also may not be specific to the cellranger version, given that this is also coming up for me in the multisample tutorial code.

Were you able to get this to work? Is there another workaround for this?