Paired End Merging

[TonyBrooks]

We are going to undertake some sequencing where our insert size is less than the sum of our read-lengths (paired end overlap).
We would like to merge our paired end reads to create single "psuedo-reads" with high quality all the way along.
A quick Google has shown me there are lots of software available for this

http://thegenomefactory.blogspot.co.uk/2012/11/tools-to-merge-overlapping-paired-end.html

I've had a play with some dummy data using FLASH and PANDA-Seq and the results seem slightly different between them (different number of final read number and different distributions) despite using the same parameters.
Has anyone done a comparison of these packages before, or have a particular feel for what they think is the best performer?

[suryasaha]

That blog is a great reference. Compared fastq-join, FLASH and read-linker from CD-HIT toolkit. Consistent results but fastq-join came out fastest testing with 5% and 20% error tolerance. Been using it since.

[Coltom]

Utility of merging PE reads?

Is/are there practical/bioinformatic reasons why merging overlapping reads isn't/hasn't been performed more often? As a 'beginner' to NGS interested in applying it to WG shotgun metgenomics it would seem the obvious thing to do. I plan to use the EBI metgenomics-InterPro portal which analyzes non-assembled sequences but wouldn't it be useful to input combined overlapping PE 'virtual' reads?

[suryasaha]

One practical reason is that the paired reads never come out with all bases the same quality so after trimming poor quality ends a overlapping 150x150 PE read can become non-overlapping (say 125x140). Therefore most tools are designed to accept the Fwd and Rev for a PE read separately along with the expected insert size. The insert size, you will discover, also varies. So keeping all these uncertainties in mind, its better to work with Fwds and Revs separately. If you can overlap some of the reads, more power to you.

[JackieBadger]

I use FLASH and it's spot on for >99% of sequences

[Coltom]

One practical reason is that the paired reads never come out with all bases the same quality so after trimming poor quality ends a overlapping 150x150 PE read can become non-overlapping (say 125x140). Therefore most tools are designed to accept the Fwd and Rev for a PE read separately along with the expected insert size. The insert size, you will discover, also varies. So keeping all these uncertainties in mind, its better to work with Fwds and Revs separately. If you can overlap some of the reads, more power to you.

I thought this might be it - a combination of 'poor' sequence quality and the lack of an experimental approach able to isolate fragments of DNA within a tight size range. Isolation from agarose gel slices (whether via a manual approach or something like the Pippin system) and SPRI approaches appear popular - has anyone tried isolating from acrylamide gels rather than agarose as these have greater resolving power?

[krobison]

I thought this might be it - a combination of 'poor' sequence quality and the lack of an experimental approach able to isolate fragments of DNA within a tight size range. Isolation from agarose gel slices (whether via a manual approach or something like the Pippin system) and SPRI approaches appear popular - has anyone tried isolating from acrylamide gels rather than agarose as these have greater resolving power?

For at least some papers, the agarose method produces remarkably tight library distributions, though when I looked at this it was when much shorter insert sizes (~120) were common; with the long reads now it makes sense to go to longer inserts, and perhaps the extreme precision isn't doable there.

Anything that adds a step may be skipped by some. Also, there is always a risk of paired end merging getting things wrong when dealing with repeats in the same size range as the overlap.