What effects will caused if mark duplicate reads before stringtie counts reads?

Question

[MatthewP]

We know gatk MarkDuplicates works on insert level for paired-end sequencing. So I think if some reads marked as duplication will have higher probability it's real PCR duplication rather than high expression transcripts. If we mark those reads before stringtie, maybe counting result will be very different? Example under different duplication rates 60%, 40%, 20%, 10%...