Removing Duplicates Scenario Exome Resequencing

[Hkins552]

I am rethinking my well-established pipeline and trying to tweek it a bit and came across some controversy on the use of remove duplicates. I would like to propose an example scenario and my current understanding of Picard Tool's Mark Dups:

At a given 5' loci (chr12:144230), two alleles are present (assume these are the first three nucleotides of 90bp reads):
1) GAC #reads = 30 Highest Q score= 25
2) GAA #reads = 8 Highest Q score= 26

It is possible that some of the 30 reads of allele 1 are due to strand bias or pcr dup. and if visualizing this on an aligner, I would have some suspicion that the locus is heterozygous.

If Mark Duplicates: Allele 2 is kept due to highest Q score, and thus there is loss of heterozygosity

If no duplicate removal: Both alleles are kept, but the ration is not close enough to 50% to be called a het.

Is my very rudimentary understanding of this tool correct and, if so, are there any other solutions?
Does anyone have recommendations based on the fact that this particular project is exome resequencing at 50-75x average depth?)

[Heisman]

With 75x coverage and I'm assuming paired end reads that has a size distribution with a standard deviation of at least 30bp, it is very unlikely for you to get multiple reads to pile up at the same exact locations without them being PCR duplicates. I would remove them for SNP calling.