effective mismatches allowed while demultiplexing

[a_mt]

Hi all,

What is the minimum/optimum number of mismatches to be allowed during demultiplexing ?? How do we decide this?

I am demultiplexing using fastx-barcode splitter, allowing 4 mismatches (my barcodes are 13mer), but I am getting large amount of unmatched sequences. When I raised it to 7, it gave decent results.

Now how do I decide or is there any criteria on which which to decide this number ?

Thank you.

[maubp]

This answer would depend on how long the barcodes are, and what other barcodes are being used on the same sequencing run. If you are multiplexing 96 samples, you have to be quite strict with the matching compared to if your only had 12 samples. The point is if you allow too many mismatches there could be more than one possible barcode matched.

I would probably be cautious and use a strict mapping. If you relax this, yes you get more reads, but there is more chance of mis-categorising them.

[a_mt]

My barcodes are 13mers and yes it is multiplexing of 96 samples. Guess I will use strict mapping.

I have another question though. After demultiplexing I got some samples with very few common paired end reads. Instead of aligning these little number of paired reads, is it valid to map all forward and reverse reads like a single end reads and merge their sams later? My main purpose is variant calling.

[Heisman]

What is the edit distance of your barcodes? Meaning, do you need there to be 3 mutations to possible convert 1 to another, or 2 mutations, or 4 mutations, etc?

You can align paired end reads as single end, but why not just align them as paired end?

[a_mt]

I don not have much information on edit distance. But it seems to be around 6-7, and barcodes are 19mers (sorry a typo at the beginning, it isn't 13).

I want to align them as single end, because there are very little number of forward and reverse reads which form pairs (around 30%), so I am loosing majority of reads and hence coverage/depth. But I was not sure whether is it valid way to do so..

[luc]

....there are very little number of forward and reverse reads which form pairs (around 30%), so I am loosing majority of reads and hence coverage/depth......

Is this because you are de-multiplexing them separately? If so, better don't do it this way.
A smart tool would use the best match information from either the forward or the reverse read.

[a_mt]

Yes.. I am demultiplexing them separately..

If you don't mind, can you suggest me any such tools you might be familiar with, performance wise... There are lot of such tools (1,2) and I don't want to play with all of them..