float vs. real in .mtx files?

[bugneville]

Importing float .mtx files (e.g., like those from multimapping) in R for downstream analysis is currently not straightforward. Would there be any limitations if STAR just output these files as real instead?

[alexdobin]

Hi @bugneville

did you mean integer instead of float?
When multimappers are included, the counts indeed become non-integer, as one read can be "distributed" between multiple genes. You can round them to integeres, but is not the best approach, I believe - see the discussion here: #1262 (comment)

Cheers
Alex

[bugneville]

@alexdobin, apologies for not responding sooner.

No -- i meant real instead of float. It appears if you simply change the header in the .mtx from float to real, the .mtx files can be read into R by the various packages (most of which appear to depend on the Matrix R package?). Is there any limitation with only real precision vs. float that you can imagine?

[alexdobin]

Hi @bugneville

you are right, we indeed need to use "real" not "float" in the matrix market header.
I will fix it in the next release.

Thanks!
Alex

[bugneville]

thanks @alexdobin!