[obenauer]

We have been having multiplexing problems too, with reads matching one influenza strain showing up barcoded as a different strain. After months of troubleshooting, I found the explanation of indexes being mis-assigned due to mixed clusters on the flow cell helpful, as found in Kircher et al., Nucleic Acids Research 2011, 40:e3, "Double indexing overcomes inaccuracies in multiplex sequencing on the Illumina platform".

You can filter all the reads, whether single reads or paired ends, by their index base qualities to reduce sample-to-sample contamination. I have programs for this, filter_multiplexed_reads.pl and filter_multiplexed_pairs.pl. I don't know if it's possible to post them here, but send me an email if any of you are interested (john dot obenauer at stjude dot org).

A major disadvantage of filtering this way is that you throw a lot of good data out with the bad. In a recent MiSeq run, filtering removed 25% fo the reads. Older data from the Genome Analyzer 2 was even worse, with 90% or more of the reads being removed.

An alternative we have been trying is sparse-matrix double barcoding, described in that same Kircher 2011 paper. Using the 96-plex Nextera XT kit (catalog FC-131-1002), we use only 24 (or fewer) of the barcode combinations at a time. This way, if mixed clusters cause mis-assignment of one of the two barcodes, it will create an invalid barcode pair that the Illumina software removes by default. We're still evaluating this, but the hope is that it will clean up our multiplexed data.

[leleon]

You can filter all the reads, whether single reads or paired ends, by their index base qualities to reduce sample-to-sample contamination. I have programs for this, filter_multiplexed_reads.pl and filter_multiplexed_pairs.pl. I don't know if it's possible to post them here, but send me an email if any of you are interested (john dot obenauer at stjude dot org).

Dear obeanuer,

Do you still can share these scripts?
In that case, do you have any other email?
Thanks in advance

Luis

[Brian Bushnell]

I wrote a program for filtering reads by barcode quality, available with BBMap. It requires you to have the barcodes as a fastq file, and is run in two steps:

mergebarcodes.sh in=reads.fq bar=barcodes.fq out=merged.fq

filterbarcodes.sh in=merged.fq out=clean.fq maq=15

...where "maq" is the minimum average Phred quality of barcodes to retain. In my testing, removal of about 25% of the reads reduced cross-contamination by about 50%, which is not great but not terrible either. Discarding chastity-failed reads and reads with barcode mismatches seems more important.