In my code I included the following lines:
library(unix)
rlimit_as(1e13)
This is supposed to increase memory limit allocated by R. Although I’ve read somewhere that the latest versions of R allocate memory dynamically and render these two lines obsolete.
I was using supercomputer provided by Compute Canada, more specifically Beluga. The peak memory use for me was 450G, but I fear that the memory usage would increase by a large margin for the dimension reduction steps.
So indeed I would also recommand subsetting the data by regions or cell types.