doublet rate after demultiplexing by cellranger

[d-lashgari]

Hi Dev Team,

Thank you for the tool; it's very useful. I have a question regarding the doublet rate. I am working with a dataset that includes multiple samples, each consisting of 7 different conditions multiplexed by 7 hashtags. The dataset has been processed with CellRanger and is already demultiplexed. However, I suspect there are doublets that were not detected by the demultiplexing process, specifically those that share the same hashtags.

For each sample, I have merged the demultiplexed conditions, as these were originally in the same well together, and I plan to run scdblfinder on this merged data. I believe I need to adjust the doublet rate, assuming that after demultiplexing, only 1/7 of the doublets remain. I calculate the adjusted doublet rate as 0.00143 * ncol(data) / 1000, based on the 0.01 rate per 1000 cells, divided by 7.

Could you confirm if this approach is correct?

All the best

[plger]

Hi,
I think your reasoning is correct, the only possible issue is that the number of cells relevant for the calculation of the effective rate depends on what was input into a single capture, rather than the merged one. Assuming that your samples are relatively homogeneous for the 7 hashtags, you should be able to use the average number of cells per capture as the basis for your calculations.
Given the uncertainty, you can also increase the dbr.sd argument, which gives the uncertainty of the dbr (by default it's 40% of the dbr), and will result in more weight being given to the optimization of the classification accuracy vs the expected number of doublets.
Best,
Pierre-Luc