Impute Gene Score matrix per chromosome

[jindalk]

Hi, I am trying to run imputation on a large dataset and have been running into memory issues. One work around I have thought of is to obtain GSM matrices by chromosome, impute them in parallel and then merge them with cbind. Is that equivalent to running imputation on the entire matrix? I was just not sure if ArchR performs any scaling as part of the imputation function which might change my results if i am running it only on a subset of features at a time.

Thanks

[rcorces]

Once you run addImputeWeights(), the weights are stored per cell in your ArchRProject. I believe those weights could be applied using imputeMatrix() on matrix chunks without any problem. You could check by testing on a single gene.

[jindalk]

Great! Yes that's exactly what I tried, loaded chunks of the GSM matrix (using the seqnames argument) and then imputed them using imputeMatrix()

Thanks