destripe function to produce 8um bins

[YuyanRuan]

Hi, thanks for the wonderful tool!

I am wondering whether destriped 2um bins after applying the function destripe will also be applicable to produce 8-um bins. Will this produce better-quality data compared to the raw 8um bins from the space ranger? Do you have any insight regarding this issue?

Thank you for your time!

[nadavyayon]

Hello, thanks for your question.

The quick answer is that we don't know how this will affect the 8um bin level. like any normalization process, it can have positives and negatives in certain situations (extent of striping, depth of sequencing, size and shape of tissue... etc)

I would recommend to try with and without destriping of your data both in cells or 8um. Our internal tests showed that the results for cells are very similar.

However, to generate the 2ndary labels, destriping is really useful as it generates a "homogenous image" that is far better than the original pattern for segmentation. We specifically have the option to generate this image without altering the counts.

Hope this helps!

[YuyanRuan]

Hi, thanks a lot for your reply and sharing!

I have two questions.

1. The 8um bins by default (square_008um) are generated by combining raw 2um bins. However, we can generate again with the destriped 2um bins provided by this tool. Have you tried this, and what do you think?
2. You mentioned internal tests. Could you please share the criteria you used? This is very interesting and useful!

Thanks again!

[ktpolanski]

We haven't tried the 8um bin reconstruction post destripe, but as you say it can be done based on just collapsing based on the array coordinates. That, or it seems that the new spaceranger outputs an additional file which carries the correspondence of various resolution bins.

A call can be made in downstream analysis based on how the data is performing. We have a notebook showing a comparison of celltypists on demo data both with and without destriping. In a practical use case in the lab, a decision between destriping and no destriping was made based on the tightness of the clustering. Like Nadav mentioned, you would still want the destriped totals for the purpose of generating the GEX image for segmentation, but you can accomplish that without modifying the expression data by running the destriping function with adjust_counts=False.