bwa mem with paired/unpaired reads

[mchaisso]

I have a quick question for all bwa mem users (or author, if you have time Heng).

I have some reads from a simulated exome study that have been modified to have a 50/50 allele balance with one allele having a 3 bp deletion at a particular locus. The reads that cover this locus are paired, and so I have a set of reads:
reads.f.fq, and reads.r.fq

If I align using the paired reads:
bwa mem ref.fa reads.f.fq reads.r.fq ...

and process the results with samtools mpileup, no variant is found.

However, if I concatenate the read files together, and align without pairing:
cat reads.f.fq reads.r.fq > reads.fq
bwa mem ref.fa reads.fq ...

the variant is discovered as expected.

Just in case I had the orientation of the reverse reads incorrect, I tried the first command with the reverse complement of the read.r.fq file, without success.

Is there any bwa mem output I should look for that would indicate improper mate-pairing or other invalid input?

thanks,
-mark

[dpryan]

Just look at the output BAM in IGV.

[mchaisso]

Just look at the output BAM in IGV.

Good comment - it is often good to review the sanity of the alignments in a viewer. I neglected to mention that the indels appear in both the paired and unpaired alignments in IGV, as well as bwa aln based alignments. Samtools mpileup detects the indel with the bwa aln based alignments.

-mark

[dpryan]

One thing to keep in mind when you treat them as unpaired, is that your coverage doubles. That could make the difference.