Background subtraction

[interactivereport]

Is it needed? If so, how to perform such operation on the count matrix, either based on expression of negative probes or something else?

[patrickjdanaher]

Great question. Short answer, perhaps to become a short article later: We don't background-subtract the count matrix, for two reasons:

1. There just aren't enough negprobes to get an accurate estimate of each cell's background level. If there are 20 negprobes, then a cell's total negprobe counts are binomial distributed with n = 20, p = the background rate, usually ~0.05. This is just too noisy.
2. Subtracting would turn our sparse counts matrix into a dense matrix, which uses too much memory.