How much phiX to spike in for ddRADseq library on HiSeq?

[AaronReedy]

Our lab is preparing for a large scale ddRADseq project (~3000 samples). In doing the calculations for how many lanes of sequencing we will need to achieve our desired depth of coverage I am finding it difficult to get a consensus answer as to amount of PhiX spiked into a HiSeq lane.

When we spoke with Illumina directly, they recommended we spike 10% PhiX into our final libraries before sequencing. They also said that in the past more PhiX was needed for the older software to correctly identify clusters. When we talked to a sequencing provider they stated that they spike in 30% PhiX for ddRADseq libraries. This seems like a big discrepancy to me.

For folks with recent ddRAD experience, how much PhiX did you spike into your library?

[luc]

Which HiSeq model are you running?

[AaronReedy]

Which HiSeq model are you running?

Originally we thought the HiSeq4000 would offer us the most reads for the money. However the provider that we talked to said that after spiking in the PhiX we could expect 180-200 million reads on the HiSeq4000 and 140-150 million reads on the HiSeq2500. This makes it seem like the 2500 is a better choice for our ddRAD libraries. Does this sound reasonable to you?

[Number6]

We typically run ~10% phiX for ddRAD libraries on our HiSeq2500 running v4 chemistry, and they've been running fine. We typically see ~91% PF (%) at ~850 K/mm2 density, ~236 M Reads, ~215 M Reads PF, ~9% alignment to phiX (these are averages from the last three libraries/lanes we've run). I'm sure all ddRAD libraries are not created equal, though. These are from a group with considerable experience making their libraries.

[luc]

We spike -in 20% PhiX on the HS4000.

[SNPsaurus]

I would have thought that having a patterned flowcell would reduce the need for PhiX to discriminate clusters since there should be fewer "almost merged" clusters. Any ideas why there might be more PhiX needed?

[luc]

Cluster identification is obviously not the problem?
I assume Illumina just ported the X10 software (designed for genomic samples only). The software is one generation behind the one of the Hiseq 2500 and the Miseq and I guess sensitive to overall fluorescence intensity changes?

I would have thought that having a patterned flowcell would reduce the need for PhiX to discriminate clusters since there should be fewer "almost merged" clusters. Any ideas why there might be more PhiX needed?

[Jean-Rene]

You should not need 30% PhiX as long as the flow-cell is not being overclustered and your barcodes have good nucleotide diversity. Our sequencing facility usually uses 10% PhiX. It has always worked for us except on one occasion where we submitted samples with very low-diversity barcodes, so I would definitely avoid that. Otherwise, 30% sounds excessive and you might lose on coverage. Do you know if the flow-cells they will be using are patterned?