How many reads will be removed after markduplicates in general?

Question

[zau saa]

Hello! Now I'm dealing with bulk DNA-seq data.

Firstly, I aligned it to human genome by bwa-mem.

Before MarkDuplicates by picard with only required options(I, O, M), its mean depth of total region is 34.

However, after Markduplicates, its mean depth of total region is only 18! Moreover, after MarkDuplicates, its bam file is only 1G smaller. Why has the depth descreased so much? The depth is calculated by mosdepth (mosdepth -t 4 -n).

[Pierre Lindenbaum]

Why has the depth descreased so much?

run samtools flagstats to count the number of marked/total reads.

[GenoMax]

Did you simply mark or actually removed the duplicates? BAM file could have shrunk if reads simply got rearranged.

[Pierre Lindenbaum]

or the compression level has changed ...

[zau saa]

I realize that I only identified the duplicates.

[zau saa]

If I forget to remove duplicates, will it influences the SNPs called by gatk HaplotypeCaller?

[Pierre Lindenbaum]

will it influences the SNPs called by gatk HaplotypeCaller?

No. https://gatk.broadinstitute.org/hc/en-us/articles/360037225632-HaplotypeCaller

These Read Filters are automatically applied to the data by the Engine before processing by HaplotypeCaller.

NotDuplicateReadFilter

[zau saa]

Thanks for your answer!

[zau saa]

Returning to the original question， would you think it's strange that the depth of bulk DNA-seq data decreases too much after markduplicates?