Could anyone inform me about the best approach for merging 2 technical replicates from the same 10x cDNA library, but run on different sequencers? Since the same cell barcodes are detected across both runs for many reads, I would like to merge them together while still accounting for batch effects.
Current approaches with Seurat, etc. assign new names to duplicated cell barcodes while merging counts or normalized data, treating the sample as though it was derived from two distinct cDNA libraries. This seems less accurate and reduces the actual read depth per cell for experiment. If the samples were run on the same instrument, I would otherwise just merge the two matrices and sum counts from identical barcodes (assuming that the technical variation was minimal).
Thanks for the help!
If you were calling variants, I might worry about slightly differing basecalling biases between different instruments, but as far as assigning reads to genes, I don't think that would be impacted much. I'd run the 10x pipeline with both fastqs together.
But I guess if you wnat to be sure, run them separately, and see how they look in Loupe when you aggregate them.
Thanks for the suggestion @swbarnes2. After checking out the cellranger aggr webpage, I found the following which was very informative:
cellranger aggr is not designed for combining multiple sequencing runs of the same GEM Well. For that, you should pass a list of FASTQ files from multiple sequencing runs of the same GEM well to the --fastqs argument of cellranger count.
So I'll give that a go and we'll see what happens.
Log in to answer this question.
It would be useful to add if the runs were done on same chemistry/sequencer type or different. e.g. both runs were on NovaSeq or were they run on two different types of sequencers e.g. NextSeq and HiSeq.
Thanks for the note. To clarify, both of these were run on a NovaSeq. However, they were run at different universities across the US.