TypeError: ufunc 'isnan' not supported for the input types

[jazberna1]

TypeError: ufunc 'isnan' not supported for the input types

[cmdoret]

Hi Jorge,

Unfortunately I am unable to replicate your problem when running the same command on two separate datasets.
Note that using --min-dist and --max-dist will restrict the loop size, but this is only meaning full for intra-chromosomal matrices. Loop pattern will still be scanned on the entire inter-chromosomal matrices.

What is your version of pandas ? It works with v1.1.3, but I could issue a fix to make it compatible with other versions, if that's the problem.

Cyril

[jazberna1]

Hello Cyril,

I have matched my pandas to be 1.1.3 as yours and rerun chromosight 1.4.1. I'll let you know what happens.

Many thanks
Jorge

[jazberna1]

Hello Cyril,

I rerun with pandas 1.1.3 but still got the same error. I'll try with chromosight 1.4.0 instead.

Thanks
Jorge

[cmdoret]

OK then, since I cannot reproduce the error, I'll close it for now. Don't hesitate to reopen if you encounter this error on a different dataset. If so, it would be nice to see your python environment to find out what package is causing the issue.

Best,
Cyril

[cmacphillamy]

Hello Cyril,

I have gotten the same error. I ran chromosight on a matrix enhanced with DeepHiC and the original. The matrix that was enhanced did work, and the original failed. Is it likely that the original matrix is just too sparse to identify any loops?

Kind regards,
Callum

[cmdoret]

Hi @cmacphillamy,

Do you get the exact same error ?
I have tried running chromosight detect on an empty matrix, but this just returns "No patterns detected" as expected, so I do not think it has to do with sparsity.

I don't really know what the issue could be, but if you were comfortable with sharing the original cool file, I could investigate a bit.

Best,
Cyril

[GeorgyChistov]

Hello! @cmdoret

I got the same problem. It disappears when I use --pearson==0.1, but appears when I set this parameter from 0.2 to 0.9.
If --pearson is 1 or higher, the chromosight said "No patterns detected", as you said.

UPD:
I revealed that pearson between 0.1 and 0.2 works good. But from 0.2 and 0.9 it raises error

UPD2:
The change of resolution allowed to increase the pearson parameter, but I still get the error when it is equal to 0.5, for example.

[cmdoret]

Hi @GeorgyChistov, since I cannot reproduce the issue on my end, would it be possible for you to share the dataset on which it happens ? It is hard to figure out the cause if I can't reproduce the error.

[GeorgyChistov]

@cmdoret of course, here is command

chromosight detect 30C_15min.mcool::/resolutions/3200 CIDs_30C_15min_res3200 --pattern borders --win-fmt=npy [--pearson=0.199

I am sorry, I cannot attach the data. Here is link to it

https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSM4585137

Additionally, I have another question to you, as to the developer. How chromosight specify the size of its scanning window? For example, we have borders on the diagonal with size 3x3 bins and borders with size 20x20. Will the chromosight find both them? So does it somehow enumerate the size of scanning window?

[cmdoret]

Hi @GeorgyChistov I downloaded the data and ran the command, but could not reproduce your error.
It may have something to do with a particular version of numpy / pandas on your machine.

Since Chromosight is distributed on bioconda, a docker container is built automatically by biocontainers for each of its release. To isolate it from your system libraries, you can run chromosight from the container if you have docker installed:

docker run -v "$PWD:/data" quay.io/biocontainers/chromosight:1.6.1--pyhdfd78af_0 chromosight detect \
  /data/GSM4585137_30C_15min.mcool::/resolutions/3200 \
  /data/CIDs_30C_15min_res3200 \
  --pattern borders \
  --win-fmt=npy \
  --pearson=0.199

As for your question: The windows size is fixed to a default value (you can change it with --win-size) which we found worked pretty well for most of our use cases. Generally speaking, Chromosight should find borders larger than the windows without problem, but may miss borders smaller than the window. If you want to detect very small borders, reducing it will probably increase sensitivity, however this will also cause many false positives.

[cmdoret]

I figured out the problem @jazberna1 @GeorgyChistov this was indeed due to a newer version of pandas.
Briefly, using np.isnan on a pandas series with dtype object now returns an error, wheres it worked before.

I fixed it in 063ba46 and released 1.6.2. Now Chromosight works with newer versions of pandas and is compatible with python 3.10.