Strange pattern in pct_counts_mt from bin2cell deconvolved VisiumHD data

Question

[danielpintard]

Hey everyone! I have been working with some VisiumHD data and ran Bin2Cell on it to group the 2 µm bins into larger bins that reflect single cells. After I ran the B2C pipeline, I took my new, deconvolved spatial data object and ran a standard scRNA-seq analysis workflow. When I calculated QC metrics, I noticed this strange, sort of banding pattern that occurs in the plot (seen below). I assume this may be an artefact of the B2C pipeline, but I am pretty lost as to what could cause this. I just have never seen a pct_count_mt plot like this before! Would love to hear others ideas and hypotheses!

[danielpintard]

Update: take a look at the plot below

Answer 1

[Kevin Blighe]

Hi!

The banding pattern in your pct_counts_mt plot is likely an artifact from the discrete nature of count data after Bin2Cell aggregation. Visium HD bins have low individual counts, and when grouped into cells, the mitochondrial reads (often sparse) result in small integer values for mt counts. This causes the percentages to cluster at specific intervals, appearing as bands in violin or density plots.

Check the distribution of raw mt counts per cell using something like hist([e-mail]$nFeature_mt). If most cells have very low mt UMIs (e.g., 0-10), that explains the stepping effect, especially if total UMIs aren't high enough to smooth it out.

It's not uncommon in spatial data with binning—standard scRNA-seq usually has higher depth per cell. If it's bothering downstream analysis, you could filter cells with extreme low counts or adjust plot bandwidth in VlnPlot.

Hope this helps.

Kevin