Techniques for analyzing and quantifying sample bleed through in genotyping with Illumina

Question

I am looking at the presence of viral genotypes within individual samples within an assay. Often times there is a sample whose read counts are firing off the charts and this sample tends to "bleed through" to the other samples.

I have recently tried confronting this problem with chi square analysis of conditional probability. For instance let's say I observe with 10 sample viral genotypes A and B. I then calculate the expected number of samples having both genotypes A and B from the number of A positive samples and number of B positive samples.

So far this is all I have. I do not know if anyone else has any ideas on how to determine and quantify bleed through in viral genotyping.

Slightly more information about this situation: these samples were run on Illumina HiSeq, we are planning on doing another HiSeq run with even more samples shortly so would like to take proper precautions with that, additionally the current pipeline utilizes Novobarcode as a demultiplexer so if anyone has suggestions on what I can implement during processing to minimize the effect of cross contamination that'd be helpful.

Answer 1

Are you using dual-index barcodes with different barcodes at each end? There is a known phenomena of "index switching" that occurs in Illumina reads. One way to control for this is to add in unamplified water samples that have unused barcode combinations where each of the indexes is shared with another sample.

More information can be found on this SeqAnswers thread. Here are some useful associated links:

• Signal-spreading preprint -- detailed investigation of the phenomenom, including methods, graphs, and associated data
• Crossblock -- tool by Brian Bushnell that attempts to remove cross contamination
• Summary by James Hadfield -- mentions a few fixes suggested by Illumina, many centred around not using barcode combinations that could conflict
• Illumina's quick-fire response -- they've known about it for a long time, and point out that it shouldn't be a problem for most purposes

I have suspicions that at least some of the index switching is being caused by ligation of two separate barcoded fragments during the sample prep (which definitely happens during sample prep for nanopore), but I haven't done any experimentation to work out if this is also the case for Illumina (due to money/time/goal constraints). The experimental design would be to take a prepared Illumina library just prior to sequencing, ligate on ONT adapters, then sequence on both the ONT and Illumina machines to see if chimeric reads discovered via nanopore sufficiently matches the index switching observed via short-read sequencing.