How do variant callers deal with overlapping paired end reads?

[krobison]

Question arising from another thread in the forums; I'm posting this separately (and cross-referencing) because I want to ask the more bioinformatics and less sample-prep oriented subset.

An apparently property of at least some hybridization capture methods is a tendency to reduce the library size. As a result, with long Illumina reads the paired end reads may overlap in the middle.

How do the variant callers out there handle this? If a variant is found in the overlap & the two reads agree, then that is clearly stronger evidence that a given variant is present in that DNA fragment. BUT, if you are worried about PCR (or sample damage such as FFPE) artifacts then you may want some separate accounting for having actually seen the same variant in two different fragments.

[nilshomer]

Question arising from another thread in the forums; I'm posting this separately (and cross-referencing) because I want to ask the more bioinformatics and less sample-prep oriented subset.

An apparently property of at least some hybridization capture methods is a tendency to reduce the library size. As a result, with long Illumina reads the paired end reads may overlap in the middle.

How do the variant callers out there handle this? If a variant is found in the overlap & the two reads agree, then that is clearly stronger evidence that a given variant is present in that DNA fragment. BUT, if you are worried about PCR (or sample damage such as FFPE) artifacts then you may want some separate accounting for having actually seen the same variant in two different fragments.

The short answer is it treats both ends as if they were from independent fragments.

What is more powerful/accurate, having two observations of the same DNA fragment (read pairs overlapping), or two observations of the same haplotype (sampling with two fragment reads)?

The former may improve the consensus call of the read, but does not increase the number of times an allele is sampled. The latter is independent of sequencing error artifacts from the same fragment and provides an independent observation of the same haplotype. Basically, if you need 30x diploid coverage to guarantee to see both alleles of a heterozygous variant, the former should count only once while the latter should count twice.