Input for ambient argument in `hashedDrops` function for 10X single cell data

[tim-meese]

Hi,

I am working with 10X single cell data and have access to both the raw_feature_bc_matrix and filtered_feature_bc_matrix for RNA and HTO. I would like to use the cells that are called by Cell Ranger. Could you please clarify what the input for the ambient argument should be in the hashedDrops function?

From my understanding, the ambient argument is a numeric vector specifying the relative abundance of each HTO in the ambient solution. If ambient is set to NULL, the function will infer the ambient profile based on the data provided1.

I read in OSCA Advanced that if we are dealing with unfiltered data, we can improve the inferences by defining the ambient profile beforehand based on the empty droplets. This involves summing the counts for each HTO across all known empty droplets, marked as those libraries with NA FDR values in the emptyDrops() output. This confuses me because I am not sure if this sum is the same as relative abundance.

Any guidance on how to correctly specify the ambient argument would be greatly appreciated!

Thank you for your clarification!

[LTLA]

Dredging my memory, it seems that the process is described in ?ambientProfileEmpty. Basically we take the summed profiles of low-count HTOs and feed this into edgeR::goodTuring. The second step is only necessary to convert zero counts into non-zero abundances, otherwise we get undefined values when dividing by the abundance vector.

If you don't have any zero counts, I suppose you could just divide your ambient count vector by the total and use that as ambient=. In fact, I suspect that the ambientContrib* functions only care about the relative values within the ambient count vector - I don't think they rely on the fact that the abundances need to sum to 1. So you could just feed the ambient counts to the function directly, though I would need to stare at it more to confirm.

[tim-meese]

Thank you for your detailed response. Based on your guidance, I took the following steps:

1. I verified that my RNA and HTO count matrices have the same unfiltered barcodes in the same order.

2. I identified putative empty droplets by selecting barcodes with fewer than or just 100 total UMI counts of the unfiltered RNA matrix.

3. I used the ambientProfileEmpty function to estimate the ambient profile of the unfiltered HTO matrix, providing it with the unfiltered HTO matrix and using the indices from the previous step as input for known.empty.

4. Finally, I used the filtered HTO matrix along with the ambient profile to demultiplex the hashes.

From what I read in the documentation and your answer; this approach seems sound.

Thank you again for your assistance.