ATAC fragment input

[WanyingX]

Hi,

I am wondering whether it allow use peak vs cell matrix as input to predict E2G pairs?

Thanks

[WanyingX]

Or can I convert my peak vs cell matrix to a binary matrix as same as format of atac fragment file?

[613weilin]

Hi, thanks for your interest! Currently, we do not support ATAC data input in the peak vs cell matrix format. This is because our pipeline relies on the ABC pipeline, which calls peaks and defines cis-regulatory regions using the ATAC fragment file.

[WanyingX]

Hi Weilin, thank you for your replies! I am testing scE2G on my datasets. Does the Pairs.Kendall. calculation usually require very long time?

[613weilin]

Hi, has it finished running now? How many cells are in your dataset? For reference, it took about 5 mins for a dataset with 7.8k cells. However, it could take longer for datasets with more cells.

[koefoeden]

Hi - just chiming in here. We also have issues with long kendall calculations, which took ~2 days for a cluster of 150k cells (which is also arguably very large) for that single step.

Do you have suggestions on how to speed-up or parallelize this step? Or is it simply grossly overkill to process this many cells at once?

[613weilin]

Thanks for the feedback! We will consider enabling parallel computation to speed up this step.

[koefoeden]

Just to be completely clear, it is the compute_kendall rule that is slow for us!

(Although the create_neighborhoods is also ~4 hours).

[WanyingX]

Hi Weilin,

A question about how scE2G corrects the ABC score equal to 1, these interactions, which are one promoter locus? Would you give a penalty score for it?

Thanks

[613weilin]

Hi @WanyingX,

Thank you for reaching out.

We assign an ABC score of 1 to elements that are promoters of the target gene (as indicated by the "isSelfPromoter" column being "True" in the prediction results).

We have not applied any penalty score in these cases. It makes sense that the promoter of a target gene would consistently regulate the expression of that gene.