1) When Picard MarkDuplicates with the REMOVE_DUPLICATES set to TRUE, there will be no duplicates written to the output file but unique paired reads will retain in the output file? Is this correct?
2) if (1) is correct, what is the selection criteria for the unique paired reads compare to other duplicate paired reads?
It associates a score to each read based on the amount of bases with qualities > 15. The one with the best scored is picked.
For more details read MarkDuplicates.java in the picard repo. In addition, let picard flag the reads instead of removing, then look at the read fiels of a set of duplicates to confirm all this.
Can you please close this one? I am pretty sure I gave you the answer you were looking for.