Multiplexing: minimum numbers of index per lane on the flow cell

[kenkclam]

Hi all,

I heard that we cannot do only 2 samples (2 barcodes) in one lane of the flow cell, and that it is better to have 6 samples per lane. Is it true?

The explanation I got is that if there are only 2 indexes, half of the clusters will give the same signal, when the sequencing reaction reach the barcode area. And this will create problem for the machine to identify individual clusters.

I am quite confused about this issue.
I would appreciate it a lot if you can share your experience/thoughts on this in layman's terms

Ken

[Heisman]

There are 4 different bases. Think of the flow cell as a rectangle with a lot of dots on it. Each dot represent a cluster. Each dot is the color that is associated with the base currently being sequenced (4 bases possible, 4 colors). So, with an equal distribution of A/C/G/Ts, there will be a mosaic of 4 different colors across the rectangle. If every successive base is independent of the previous base, then two adjacent clusters with an A on the current cycle will have a good chance of having different bases on the next cycle. If clusters are determined by a sequence of 4 bases, and there was a 1/4 chance of getting any base, then there would be a (1/4)^4 = 1/256 of two adjacent clusters having the same sequence of bases. So, that's pretty small.

If you have two barcodes only, and half of the fragments have one barcode and half have the other, then there would be a 1/2 chance of two adjacent clusters having the same sequence of bases. If these are close enough, the imager will not be able to distinguish between them, and will not call them separate clusters.

Does that explain to you why you need more than 2 barcodes?

[GenoMax]

But in practice two barcodes per lane will work fine as long as you do not overload the samples.

[henry.wood]

Whenever we want to run a small number of samples on one lane, if we have enough DNA we make more than one library for each sample. If DNA is limited we do the steps up till tagging as normal, then pool 2-3 tags into one tube and tag the samples with the mixture of tags.
You then just sequence as normal, demultiplex, align and merge the relevant bam files back into on file.

[TonyBrooks]

There's an Illumina tech note somewhere on which barcodes to use for low level pooling using TruSeq adapters. It just went up on iCom earlier this month.

Look for
TruSeq_SamplePrep_Pooling_ProductInsert_15034135_A.pdf

Basically, you want to balance the red (A/C) and green (G/T) lasers where possible.

For 2 samples, use AD006 pooled with AD012 or AD005 pooled with AD0019. These are from Set A only. There are no recommended combinations from Set B adapters.

[pmiguel]

[...] If clusters are determined by a sequence of 4 bases, and there was a 1/4 chance of getting any base, then there would be a (1/4)^4 = 1/256 of two adjacent clusters having the same sequence of bases. So, that's pretty small.

If you have two barcodes only, and half of the fragments have one barcode and half have the other, then there would be a 1/2 chance of two adjacent clusters having the same sequence of bases. If these are close enough, the imager will not be able to distinguish between them, and will not call them separate clusters.

[...]

So this seems to be an explanation of why you can't have low sequence diversity in the first 4 bases of read1. But index reads are done during read2 -- cluster calling has already been done at this point.

Near as I can tell, low sequence diversity cycles in an index read is not an issue on the HiSeq. At least, our HiSeq does >95% "demultiplexing" of single index lanes.

GA-IIx, I don't know.

HiScanSQ -- there is an issue that makes it difficult to use a low number of indexes in a lane. Or it did, recent changes in HCS may have improved the situation. The issue for the HiScanSQ is that it does scanning in 2 passes, one A/C detection and one for G/T detection. It was the case that if the instrument could not detect any signal in a panel during a scan, it would sometimes decide to adjust its focus. This adjustment would defocus the panel and no more useful data would be collected from that panel.

Generally, if we chose combinations of indexes that gave both and A or C and a G or T for every cycle of the index, it would work okay.

I think the MiSeq may be like the HiScanSQ, but am not sure.

--
Phillip

[kenkclam]

Thanks all of you! Very helpful replies.

So, it seems not to be a problem anymore for the HiSeq, which I am going to use. I will just go ahead with index combination suggested by Tony, or use multiple adaptors for one single sample.

Thanks again, good luck to all of you!