Anyone have any idea why this ecperimental design is prompting a "Design matrix not of full rank." error?

Apparently "lane6" and "lane11" are the problems. lane11 only has a single member which I thought might be the problem, but giving this sample membership of lane10 insteap of 11 doesn't help, it just shifts the problem to "lane7" and "lane10".

Easy. batch is equal to the sum of lane5 to lane11. So, for any estimate of the batch coefficient, you can obtain an equally good GLM fit by increasing batch and decreasing all lane5 to lane11 coefficients by the same amount. To fix this, you should drop batch from your design, as the batch effect is captured by the lane coefficients anyway. Note that if you block on the lane, your last sample is basically useless, because all its information is used to estimate the lane11 coefficient.

That said, is it really necessary to block on the individual lanes? For current (Illumina) protocols, I'd be surprised if sequencing lane had a large effect. If the samples don't cluster by lane on a MDS plot, I would forget about blocking on the lanes and just block on the batch effect, you'll get more power with a simpler model. Whether or not spike-in mix is added seems like an unnecessary factor as well.

OK, I see. I was hoping to be able to probe what the impact of batch and lane were seperately rather than leave one out. I agree that lane probably doesn't have much effect, but figured I'd actually try for a full description of the experiemnt. Thanks.