Is BWA consider insert size when define read mapped in proper pair

Question

[Medhat]

When BWA assign flag 2 for read mapped in proper pair Does it take in consideration the insert size? (from searching I think yes), But in the read below I see that the insert size is 6.
So what define proper insert size? Is't small value of insert size could point to a potential deletion? (like the case below)

HISEQ1:19:H8VDAADXX:2:2106:8906:10302 99 1 7559182 30 57M4I75M4D12M = 7559060 6 GGTGTGATCCTGGTGTGCGTGTGTACGTATATGAGTGTGTGTGTGCATGTGTGTCCATGAGTGAGTGTGCAGAGAGGATGGTGTGATCCTGGTGTGCATGTGTATGTATTTGAGTGTGTGTGTGCATGTGTGTACATGAGTGTGTAGA @BBFDDFFHHHHHJIIJJJIJIIIJJJIJJJJJIJGGHHGHHIHIJIJJIIIJJJJJJJIIBFCHFHEHBHIHHHHFDFBC>ACDDDCCCCD;<?BCCCA>@5;@C:>>@CD4<:@@A<A<?AA>ADCDDDBDCDDCC:@CCCC2>>3 NM:i:9 MD:Z:105A26^TGAG12 MC:Z:26S18M6D104M AS:i:123 XS:i:108 XA:Z:1,+7559103,148M,9; RG:Z:sample_3

Answer 1

[Pierre Lindenbaum]

if I'm not wrong for each batch of fastq data, bwa use a percentile of the read-length distribution. see:

https://github.com/lh3/bwa/blob/master/bwamem_pair.c#L84

https://github.com/lh3/bwa/blob/master/bwamem_pair.c#L93

https://github.com/lh3/bwa/blob/master/bwamem_pair.c#L99

[Medhat]

So, The answer is Yes it takes insert size in consideration and it depend on the batch that the read belongs to (in such case 6 will be proper insert size even though it does not fit to overall average of insert sizes in file).