UMI collisions when number of original molecules is very high

[fabio-t]

We have a 12bp UMI, and our collaborators sort about 1 million naive T cells from each sample. In these cells, there can be several target RNA molecules, so the total amount of original molecules comes dangerously close to the space of possible UMI combinations (16.7 millions).

Have you found that the actual UMI space is significantly lower than the theoretical limit? I assumed not - that I could, then, rely on having negligible collisions even with more than 10 million original molecules. But I could not find a proper answer to this.

To say it another way: in your experience, what is the maximum "safe" amount of molecules one should have for a 12bp UMI? Whereby one could only expect negligible "true collision" rates.

(of course, a 14bp UMI would be better, but this is what we got for now)