At which point should I perform ambient RNA correction?

[Chaosli98]

First of all, thank you for providing such a great tool!

From my understanding, ambient RNA correction should be done before any operation that could potentially affect the cells, specifically on the raw matrix generated by 10X.
However, I noticed in the tutorial on https://www.sc-best-practices.org/, the Fabian group first removes low-quality cells based on mitochondrial RNA proportion, etc., in the quality control section, and then uses SoupX. This has left me somewhat confused. Do you think this workflow is feasible?

I really look forward to your response, as this question are crucial for many people who are struggling with different pipelines.

[GiveHeartToU]

When combined with the SoupX algorithm and personal insights, this deconvolution method can be seen as a linear operation applied to the proportions (i.e., the fraction of expression for each gene in each single-cell droplet). This suggests that manipulations, such as estimating contamination fractions or removing contamination, are performed on individual cells.

Therefore, the input TOC matrix required by SoupX could consist of data from droplets that have been filtered to exclude doublets or any other droplets that do not meet single-cell quality control criteria, similar to the approach used by Fabian. Importantly, you need to ensure that the TOC droplets are a subset of the TOD matrix (which contains all droplets). Lastly, there is no need to worry about removing the filtered cells from TOD—background expression profiles do not require this subset of data.

If there are any misinterpretations or misunderstandings, please also point them out.