Depth Normalization across samples

[piotay]

Hello all,

I really appreciate your thoughts and consideration.

We have a dataset with several samples with quite different starting loaded cells (see sample details below) and quite different resulting read depths, even after resequencing to add reads. We want to downsample the reads so that all samples have similar read depths before our analysis. We have first tried DroplietUtils: downsampleReads, but we aren't sure of its success.

What should we expect if depth normalization is successful?

My understanding is that we have two options to deal with this. We are trying DroplietUtils: downsampleReads on the molecule information HDF5 file. It says "Subsampling the reads with downsampleReads recapitulates the effect of differences in sequencing depth per cell. This provides an alternative to downsampling with the CellRanger aggr function or subsampling with the 10X Genomics R kit."

However, when I try this using, I do not see drastically similar nCount_RNA or nFeature_RNA distributions for each sample. I am able to try cellranger aggr, but I want to ensure that is actually a better alternative and that we should expect different results than downsampleReads.

Cellrangeraggr: When aggregating data from different libraries, cellranger aggr normalizes for effective sequencing depth by subsampling the reads. By default, cellranger aggrcomputes the subsampling rate for each library based on the mean number of filtered reads (identified as in cells) mapped confidently to transcriptome per cell for each library. Libraries other thanthe one with lowest values are downsampled.

Is this goal and approach rational? What would you do?