Read number variation in pooled/multiplexed samples -- any tips?

[gringer]

One of the researchers at my institute did some graphs of total read numbers for a recent HiScanSQ run (see attached files). I've seen worse, but it would be nice to know what we could do better.

Ideally, we'd like to see an equal number of total reads for each sample, but this doesn't happen because things like reality, physics, and sampling error get in the way. Does anyone have any tips on how to reduce the variation in read count across pooled / multiplexed samples?

I've only had limited lab experience (my learning has been mostly theoretical), but here are some things I've thought about that might play a role in this variation, most which I would expect to be difficult to control for:

• Aliquots for quantification pre-pooling is not representative of the entire sample (i.e. not shaken enough)
• Aliquots for cluster generation are not representative of the pooled samples (or the quantified aliquots)
• Preferential cluster generation
• Missing barcodes
• Bad spot identification on sequencer

Any other thoughts on this matter?

[GenoMax]

We routinely do 2 samples per lane for a project (hundreds of samples) and see a fairly even (more or less) distribution of reads for most lanes.

I have been told by the people in the lab that the use of Qubit assays for accurate estimation (these are RNA-Seq samples) does the trick. We do use illumina bar-codes.

[csquared]

Accurate quantitation and normalization before pooling is the key.

Anything other than real-time PCR based quantiation and you will see variability from run to run, especially as you get to higher multiplexing (above 12).

Less than 12 and you can get away with Qbit or bioanalysis as long as the relative library qualities are good and consistent.