subsample from dsc-pileup UMI file to simulate lower sequencing coverage?

[daskelly]

I would like to empirically model what happens to the demuxlet sample identity call if I were to have lower sequencing coverage. For example, how does output change if I had 50%, 55%, 60%, ... 95% etc of my data.

It would be straightforward -- though cumbersome -- to subsample the BAM file and rerun dsc-pileup and demuxlet using each subsampled BAM.

However, given that all information for running demuxlet is contained within the dsc-pileup output files, it seems it could be far more efficient to downsample from the dsc-pileup files generated on the full BAM. Is it possible to manipulate the dsc-pileup output files to simulate less sequencing coverage? It looks like perhaps randomly sampling from the UMI file output by dsc-pileup might work. Is that true? (If yes, I assume this would, strictly speaking, be sampling UMIs and not reads ...)