Error with Picard MarkDuplicates

[chrismit]

I have some targeted sequencing that I'm aligning in the following manner:
1) bowtie2 alignment to only the targeted regions
2) take the unaligned reads, align to the whole genome
3) remove all unmapped/secondary alignments from each file
4) merge using Picard

The issue is when I am doing MarkDuplicates, I have reads which are aligned in both files. I don't understand how this is possible, as the reads aligned in each case should be mutually exclusive.

[kmcarr]

I have some targeted sequencing that I'm aligning in the following manner:
1) bowtie2 alignment to only the targeted regions
2) take the unaligned reads, align to the whole genome
3) remove all unmapped/secondary alignments from each file
4) merge using Picard

The issue is when I am doing MarkDuplicates, I have reads which are aligned in both files. I don't understand how this is possible, as the reads aligned in each case should be mutually exclusive.

My guess is that in taking the "unaligned reads" after step 1 the set may have included read pairs where only one mate was unmapped, but that's just a guess.

More to the point, the workflow you have described is not really the a accepted standard for analyzing targeted sequence data. You should just align all your reads to the whole genome from the start. After cleaning up your alignment (e.g. mark duplicates, remove secondary alignments, etc.) then focus on your targeted regions. This is more correct bioinformatically and will avoid the problem you are now dealing with.

[chrismit]

You're right on the read pair, bowtie2 spits out both pairs into the aligned file if either fail to map (but keeps the alignment in the bam file as well).

The reason I did it like this was because I wanted to overestimate any possible SNVs, which for the current project a conservative estimate of non-SNVs is desired. I'm going to just switch steps 1&2 and remove ones with a partial alignment from the fq.