When can I get away without a control lane?

[lparsons]

Our users are interested in getting the highest throughput and lowest cost they reasonably can from the GA. As such, they would like to investigate eliminating the control lane from at least certain runs.

In my understanding the main purposes for the control lane is matrix and phasing estimation. If we are using genomic samples from an organism with a balanced genome, can we safely eliminate the control lane and use the default behavior of using all lanes for that estimation?

Does anyone have experience with this? Are there other things I should consider?

Thanks for your time.

[hoisinjl]

This is my first post, and something I am very familiar with. I run two GAIIx's full time and actually have never used a PhiX control lane. I would say that you can safely not use a control lane (CL) when you have a balanced sample on a flowcell. It is not necessary for the whole flow-cell to have balanced samples as you can specify --control-lane in the pipeline, I would not average across all lanes if not all samples are balanced.

Everyonce in awhile, I will be a run that does not have an adequate control-lane, and I am required to use the phasing, prephasing and matrices when a previous (or the next run). Keep in mind that this is doable, but definitely not ideal and can lead to higher error rates.

[basickler]

If you're running human or something with a GC content similar to PhiX you can ditch the control lane. Anytime you deviate too much from that GC ratio you should put a control lane in the flowcell. If you run mixed flowcells you can use a lane something with a decent GC ratio as the control so if you plan correctly you can usually get around it.

Brad

[greigite]

You might want to check out the Ibis basecaller (http://genomebiology.com/2009/10/8/R83), which supposedly has lower error rates than Bustard. They specifically say a control lane is not needed, but you do need something that can be aligned to a reference to train the basecaller:

"Ibis is not dependent on the inclusion of the PhiX control lane. In the case of resequencing projects or projects where some subset of the sequences generated comes from a previously characterized genome (for example, mitochondrial sequences) it is possible to use these data as a training dataset for Ibis. We have shown that it is possible to use the mitochondrial sequences generated as part of a shotgun sequencing experiment as an alternative training set (Figures S7 and S8 in Additional data file 1). Further, the raw Bustard output can be used as training data in cases where there is no reference set available (Additional data file 1), although the reduction in error rate is less than can be obtained when a reference is available."

[simonandrews]

I would add one comment, which is that it's OK to not use a control lane if you have a balanced, random sample with sensible GC content which you trust. The last bit is important. As well as acting as a control for phasing PhiX is also very useful in that it's a library which is known to be good.

Having a lane you wanted to use as a control fail, or even just end up full of primer would be a major pain. We still use PhiX on every flowcell, but I know of other sites who do long term projects on big libraries by making them in large quantities, testing them, and then using them as the control lane sample.

I wouldn't want to use a sample I'm running for the first time as a control lane - but maybe I'm just overly pessimistic...