[pmiguel]

Only time problems with indexes show up on HiSeq is when one over clusters samples. Regular reads are resistant but the index reads tend to start accumulating N's (when samples are over clustered) leading to losses.

What threshhold do you start to see this? We get very high demultiplex results (>95% demultiplex of PF reads) at 1000/mm^2.

--
Phillip

[GenoMax]

What threshhold do you start to see this? We get very high demultiplex results (>95% demultiplex of PF reads) at 1000/mm^2.

--
Phillip

Some point north of 1000/mm^2. The tipping point is rather abrupt (as you may have experienced). We try to stay in 900-950 range to be safe.

[colinkingswood]

I am working on software to calculate this, hopefully I will release soon.

[massspecgeek]

Hi Big_SNP,
Yes, I can help -- pool any indexes you like together. It doesn't matter any more. Illumina fixed this issue that caused low % demultiplexing due to unequal base representation around the time they got around to mentioning in the manuals it was a problem. Now we have manuals that warn against a non-existent problem.
Ah well...

--
Phillip

Just for the record, as of today (1/24/14) Illumina tech support says that there haven't been any changes made that address this issue and they still recommend having color balance in the index read whenever possible.

Roger

[pmiguel]

Just for the record, as of today (1/24/14) Illumina tech support says that there haven't been any changes made that address this issue and they still recommend having color balance in the index read whenever possible.

Roger

Roger,
Well, I can only comment as someone who did have issues with what later was called "color balance" on our old HiScanSQ. These were severe issues that would make demultiplexing either impossible, or require special tweaking of CASAVA to get them to work.

Just as a metric, any lane with only a single index in it rarely yielded more than 50% of the reads after automatic demultiplexing. Not that it mattered as there was only one sample the lane. Still, it gave and indication of how much of an issue "color balance" was.

Now, at this time, there was nothing in Illumina literature indicating this was a problem. But it clearly was. So we figured out how to balance our indexes on our own.

At some later point, color balance became much less of an issue. Again this is based on the assay "I just ran a single library in a lane, did it demultiplex?" The answer used to be "No, not very well." But it became, "yes".

Since we upgraded to the HiSeq we have been unable to detect any loss of reads due to lack of "color balance". So, I took that to mean "Illumina fixed the problem".

However, around this time, there were additions to many of the library construction protocols specifying the importance of color balance. Those sections, near as I can tell, were now completely unnecessary. If they could be sent back in time a year, they would have been great though.

Anyway, if you want to color balance your library pools, feel free. I don't think it actually yields you more reads these days, but I don't think it hurts anything either. Maybe as an aesthetic statement it retains its importance. Sort of like wearing a waist coat to carry your pocket watch.

--
Phillip

[clintp]

I have had the same experience as pmiguel--that color balance doesn't really seem to affect demultiplexing, at least on the MiSeq.

When you use Illumina's Experiment Manager to set up the run, having unbalanced indices gives a warning, but it lets you generate a sample sheet anyway.

With any run, I find that the spiked ("nonindexed") PhiX ALWAYS shows up in each demultiplexed sample, not just in Undetermined. MSR isn't going to just put any read into the demultiplexed files, it has to have an index sequence of some kind. They don't say it does, but I wonder if the Illumina PhiX actually has some low level of indexed molecules (even a randomer in the index spot) to help prevent the color balancing problem.

This is testable, but I haven't tested it. But it makes sense for Illumina to do this and gives customers a reason to buy their PhiX. AND, it actually looks pretty bad for Illumina if this is NOT the case--the index sequences do not exist in the natural PhiX genome, so if MSR is randomly assigning sequences to bins, that is a bad thing indeed.