How to use DownsampleSam quickly and efficiently with multiple PROBABILITies?

Question

[-IvanWoo-]

I want to build a standardized quality control process for particular RNA-seq samples, so I intend to downsample the alignment results (BAM files) with multiple proportions.

I tried to use DownsampleSam in Picard tools, and my command was as follows.

for SAMPLE in sample1 sample2 sample3 sample4 sample5 sample6; do
  parallel --env SAMPLE --keep-order -j 8 '
    picard DownsampleSam \
    I=${SAMPLE}/raw.sam.gz \
    O=${SAMPLE}/{}downsampled.bam \
    A=1.0E-5 \
    P={}
  ' ::: 0.001 0.002 0.005 0.01 0.02 0.05 0.1 0.5
done

This command does what I expect, but it seems to be more time-consuming and computationally intensive.

Considering that hundreds of samples may be processed later, I wonder if there is a more optimized solution? For example, when doing P=0.5, is it possible to output the results for P=0.001, P=0.002, P=0.005, P=0.01, P=0.02, P=0.05, P=0.1, and P=0.2 together? That would save a lot of computing power and time if it were possible!

Thanks in advance!

[ATpoint]

May I ask what downsampling has to do with QC? As for the parallelization, if you want to further parallelize then wrap this snippet itself into parallel rather than a loop or submit an array of jobs if you are on a cluster that supports this, e.g. via SLURM.

[-IvanWoo-]

I want to analyze the coverage of specially treated RNA by downsampling. When I do the for loop I used bsub to submit to multiple compute nodes of hpcc, but it is still a bit of a waste. I think downsampling should be able to get multiple scales of results at once.