Merging PE reads taking into account minimum positional quality score

[martinjf]

hello everyone,
I face an unusual issue with merging Illumina Paired-end reads and controling for the merging using individual base quality:
using your preferred merge (let us say Pear, FLash...) you can of course control for the effect of quality score difference between R1 and R2 for a given position and decide wether the difference is large enough to make the base with highest value the one you keep. Ok, works fine in most cases. Here I have a different situation where I would like to do that + if one of the two scores for a given position is < threshold (let us say 20 for example) then the other strand is kept, whatever the difference in score AS LONG as its own score > 20. And that I could not find it from any PE reads merger yet !
Any (verified) idea anyone?

just to avoid out of topic comments, &- yes, I alreadyy though of softmasking low quality bases before merging but I could not find any merger that uses this information also. 2- No, I cannot just remove the reads with low quality bases before merging as I cannot use a strategy based on the % of low quality reads or sliding windows as I really want to use the individual position quality profile for rare variants calling. 3- yes, using illumina correction algorithms like dada2 is an option I will also explore but I would prefer exploring the solution I detail during merging first.
Thanks all !

[Markiyan]

You can do custom trimming (based on different profile of errors between R1&R2 (to take into account the higher chance of presence of the erroneous/random data in the end of the R2 read)).

One can use perl for prototyping a read trimmer (or modify any of the open source ones (if you have a bit of C/C++ knowledge).

May I ask you a couple questions:

1. What readlength and platform had you used for you sequencing data aquisition?
Was it 4chanell or 2 chanell imaging? 2 chanell one has 2x-10x higher the error rate and is limited to 150bp read length, so use a 4 chanell platform.

2. What was your cluster density? If you use MiSeq or a Hiseq 2500 - I would undercluster to get lower error rate if looking for rare things - lower cluster density lowers error rate by 3X-6X, (while also lowering raw data yields).

3. Did you use the PCRFree library prep protocol? What was your average insert size and size distribution?