Reading in .h5ad files

[ccnawrocki]

Hello, I have been trying to use Kana to look at a scRNA-seq dataset that I have. It is in the .h5ad file format. However, when I try to read it in, I keep getting this error message:

What do I do to fix this? Any help would be greatly appreciated.

[LTLA]

Hm. Negative size factors shouldn't exist in count data, given that we derive our size factors from library sizes. This error message implies that you have negative "count" data, possibly something that's already been normalized (though it would be somewhat unusual for a normalization strategy to generate negative values).

Given that you're using explore mode, you might have some more luck if you specify that the assay is already normalized:

This will instruct kana to omit any further calculation of size factors.

[ccnawrocki]

Ok, I tried that and it worked. Thanks.

[ccnawrocki]

I had another question that is similar. When I tried to load in another .h5ad, I got the following message. This file is pretty large. Could that have anything to do with it?

[LTLA]

How large is the file?

I've never seen this error before, though I don't think it's the size. If I took the error message at face value, I'd guess that some other process is holding a lock on the file, e.g., you've got a Python session that's got an open handle to that H5AD file.

I vaguely remember that I wasn't able to open a HDF5 file in R if I was reading it in Python. Perhaps this might help.

[ccnawrocki]

The file is 6 gb. How do I close that h5ad file? Not much to go off of with that link.

[LTLA]

Okay, well, 6 GB is probably going to be too large to get into the app.

That's an awfully large file. What's inside it? If it's just a count matrix, it must be at least 1-2 million cells to reach 6 GB. The client-side paradigm can't handle that kind of volume at the moment, it would need a (yet-to-be-released) backend service to do the compute.

[ccnawrocki]

Oh okay. It is a counts matrix with a decent amount of meta data. It has been clustered at many resolutions.