Could not compute similarity for...

[vaquerizaslab]

Could not compute similarity for...

[kaukrise]

Hey @liz-is ,

thank you for the detailed bug report. Can you please try to plot the O/E matrix of a chromosome (or part thereof) that fails? I have a suspicion that the expected values might be the issue here, in which case this is probably related to the FAN-C dev version.

Thanks!

[liz-is]

Thanks for looking into this Kai! Here's the O/E matrix for the same dataset and chromosome.

[nickmachnik]

Hey Liz!
There is a lot of white in this matrix, which according to the colorbar is oe=1. Are all these values actually 1 or very very close to 1?
1 is the default masking value for unmappable pixels in chess. All 1 matrix rows are marked as unmappable rows if the row sum equals the row length (looking at the code now this already doesn't seem ideal to me). This is not done for the whole chromosome matrix, but only on the submatrices that are compared; so a row doesn't have to be all 1 for the whole chromosome, only in a particular compared region in order to be marked as unmappable.
You could try to increase the fraction of unmappable bins that chess permit with --mappability-cutoff (maybe 0.5 or even higher?). This is not a fix, but might point out if this bug has something to do with false masking or computation of oe values.

[kaukrise]

Hi @nickmachnik ,

this was an issue with the FAN-C development version, which we could figure out independently, so I am closing this!