ComputeMatrix fails to pick up a fraction of values

[eglsdl]

Hi,

Some of my heatmaps don't pickup signal on fraction of the regions. Example:

computeMatrix reference-point -S input.bw
-R regions.bed
-o matrix.gz -p 4
--missingDataAsZero
-a 1000 -b 1000

plotHeatmap -m matrix.gz
-o matrix.png
--colorMap Reds
--sortRegions no

Generates:

When I look at matrix.gz I see regions like:
chr11 120528204 120528272 55667.2 1554.25 . 0.000000 <...> 0.000000

When I check on IGV:

There is clearly signal in that region. There are a lot of regions like that spanning over different chromosomes.
I have attempted increasing allocated memory but that did not change anything. The only error I see is a downstream one, I assume occurs when calculating the means for sorting:
/util/opt/anaconda/deployed-conda-envs/packages/deeptools/envs/deeptools-3.5.1/lib/python3.9/site-packages/deeptools/heatmapper.py:1219: RuntimeWarning:
Mean of empty slice

Any advice?

[eglsdl]

I am looking at it further and computeMatrix seems to be skipping whole chromosomes 10, 11, 12, 13, 14, 15 and a big chunk of chromosome 16. The bottom row is the regions I am trying to plot, middle row is regions that in the matrix come up as 0, top row is ATAC signal.

Zooming in on chromosome 16:


I do not see the reason why some of these signal regions are picked up and some are not, the start and end of the "missing data" regions seem to be abrupt. I am not getting any error messages. I have tried using single thread, increasing allocated memory significantly, enabling/disabling sorting, using different deeptools versions (2.5, 3.0, 3.5) with no success. I am analyzing 86137 regions but allocating 80GB of memory so I doubt that is the issue. My colleagues have encountered this problem too but were not able to solve it. Any help would be appreciated.

[eglsdl]

I also attempted to plot just the regions that come up empty and they remain empty. I tried inspecting the bed and bigwig (i.e. corresponding bedgraph) by hand and I do not see any difference at all between sections in chr16 that get plotted normally vs get marked as 0.