Does providing a number of expected donors force vireo to find that many donors?

[LilianaLucca]

Dear vireo people,

thanks so much for creating such a useful tool!
I am working on a large single-cell RNAseq (10x 5 prime) dataset of human donors for whom I do not have a reference genotype. The samples are batched two-by-two, mixing a biological male and a biological female, so that by combining the expression of some sexually dimorphic genes (such as RPS4Y1 and XIST) with the donor1 vs donor0 vireo attribution, we can figure out who is who.

This approach works fairly well, but in about 10% of our 100+ batches, we are presented with cells that appear to be only male or only female, yet attributed to two different individuals by vireo. I am trying to discriminate between two possibilities:

1. that for some reason the emulsion contains only cells from a single donor, but vireo tries to attribute them to two individuals because it expects there to be two individuals
2. that there has been some clerical error in annotating the biological sex of the donors, and we do in fact have two individuals of the same sex in the emulsion.

This is the command that I have run:

cellsnp-lite -s ${input_dir}/$batch/outs/per_sample_outs/$batch/count/sample_alignments.bam
-b ${input_dir}/$batch/outs/per_sample_outs/$batch/count/sample_filtered_feature_bc_matrix/barcodes.tsv.gz
-O ${cellsnp_lite_output_dir}/$batch/
-R ${ref_dir}/genome1K.phase3.SNP_AF5e2.chr1toX.hg38.vcf
-p 5
--minMAF 0.05
--minCOUNT 20
--gzip --genotype
--nproc 5

vireo -c ${cellsnp_lite_output_dir}/$batch/
-N 2
-o ${vireo_output_dir}/$batch/

Any idea on how I could figure out where is the truth?

Many thanks!

Liliana

[huangyh09]

Hi Liliana,

Thanks for sharing the issue. Let me rephrase the issue: In some of your pools, you would expect only one donor (either you only put one donor there or you only found one sex there (by RPS4Y1 and XIST, etc), but Vireo gives two genetic clusters/donors.

Here are some thoughts:

1. if there is genuine only one donor, vireo can output two clusters if forcing it to do so. However, in this case, you may not see a big genetic difference between the two "genetic clusters" (e.g., below 0.3).
2. if there is a pool of two same-sex donors, you may see more genetic differences between them (e.g., above 0.4).

Below, is a reference for comparing genetic differences among donors:
https://nbviewer.org/github/single-cell-genetics/vireo/blob/master/examples/donor_match.ipynb

I assume you don't have the genotypes of these donors, otherwise, using their genotypes as a reference will make things easier.

Yuanhua

[LilianaLucca]

Dear Yuanhua,

thank you so much for your swift reply, and apologies for taking so long to respond.
I see that I already have heatmaps of the Geno Prob Delta in the output of the command I ran, the files called fig_GT_distance_estimated.pdf (I should mention that I am not fluent - or even a beginner - in python). Indeed, the values displayed for the donor0-donor1 pairs of the good batches are around 0.52, whereas for the suspicious batches, they are around 0.27 or less. Is that a good indication that there is indeed just one individual?

Thanks again,

Liliana

[huangyh09]

From the one example we tested, within donor has 0.15-0.22 difference between two estimation methods. In your case, you intentionally split a group cells into two sub group, then 0.27 seems in the range of the same donor-split. However, this is only based on the one example we tested in the above notebook for your reference. -Yuanhua

[DarioS]

So, it is basically akin to overclustering followed by agglomerating.