Reproducibility issue with computeGCBias/correctGCBias

[kpadm]

Hi,

I am having a couple of issues with running GC bias correction using deeptools as part of a CNV calling pipeline.

1. The DLRS scores differ slightly when number of cores used to run the GC bias correction is changed. Moreover, when the same number of cores is used, different servers give slightly different scores. The seed for correction was manually set at the same number using np.random.seed().

2. When GC bias correction is done in parallel for multiple samples, it takes significantly more time for each sample, than running one sample at a time in a serial fashion.

I would appreciate some insight into why these happen, and how I can fix it. Thanks.