SCENIC+: pycistopic QC errors

[skoturan]

Hi, I'm trying to run the new SCENIC+ using this tutorial:
https://pycistopic.readthedocs.io/en/latest/notebooks/human_cerebellum.html

When I get to the QC steps after inferring consensus peaks, pycistopic qc does not run. The error looks like this:

/.....site-packages/pycisTopic/topic_binarization.py:319: RuntimeWarning: invalid value encountered in divide mean1 = np.cumsum(hist * bin_centers) / weight1
..../site-packages/pycisTopic/topic_binarization.py:320: RuntimeWarning: invalid value encountered in divide mean2 = (np.cumsum((hist * bin_centers)[::-1]) / weight2[::-1])[::-1]

The strange thing is, it ran once already and I got some plots for sample level QC. But the TSS enrichment look very odd (attached image), and different from previous runs on the old SCENIC+ version....

When I ran the following code it seemed to work earlier:

pycistopic qc -f /rds/general/............../atac_fragments.tsv.gz -r /rds/general/................../consensus_peak_calling/consensus_regions.bed -t /rds/general/.................../consensus_peak_calling/tss_hg38.bed -o /rds/general/............../outs/qc/MCSF_16h_batch1 --min_fragments_per_cb 100 --tss_flank_window 1000 --tss_smoothing_rolling_window 10 --tss_minimum_signal_window 100 --tss_window 50

I added the the following patraemters to match the analysis I did in the old SCENIC+:
--min_fragments_per_cb 100
--tss_flank_window 1000
--tss_smoothing_rolling_window 10
--tss_minimum_signal_window 100
--tss_window 50`

I'm uncertain if they make sense or if it's appropriate for the dataset.

Any suggestions or ideas on what could have gone wrong?
The data is the single-nuclear ATAC modality from 10x multiome.

Thanks in advance!

[ghuls]

What are the values for weight1 and weight2? The error seems to indicate that there are zeros there (or infinity or some other weird values).

[skoturan]

Hi!
Thanks for your message.

This is what the values are:

Lots of zeros..

I don't understand why..also not sure if things went wrong at the peak calling step or at consensus_peaks.to_bed.

The bed file looks like this:

Any thoughts on what's going wrong?