NovaSeq Experience

[nangel]

Hi, I have seen a lot of the theoretical discussion around the NovaSeq and the resulting data. I was just wondering if there were any users that were actually running a NovaSeq that could comment on how it was performing in the lab in terms of loading, errors, general issues etc?

[Brian Bushnell]

I don't know anything about the lab issues, but the sequence quality is good. Coverage exhibits slightly more bias than HiSeq for the same libraries. Using unique dual barcodes and performing deduplication are both essential, because of the high crosstalk and duplicate rates.

[nucacidhunter]

It will be good instrument when Illumina comes up with real fix(es) for index hopping issue. Currently its use is limited to sequencing large whole genomes where available UDI combinations can be utilized and not many are willing to use it knowing that their discoveries or positive expected results could be due index hopping.

[kcchan]

Honestly this index hopping issue is way overblown. For most sequencing applications it's a non-issue. From our experience a clean library (i.e. no visible primers or dimers in BA trace), the hopping rate is less than 1%. What's causing problems are those who got used to the fact that the MiSeqs and HiSeq 2000s were able to sequencing bad libraries without issues and then running into problems with pattern flow cells. Once you understand the nuances of the new instrument it's a great platform and much easier to use compared to the older instruments.

[Markiyan]

Originally Posted by nucacidhunter

It will be good instrument when Illumina comes up with real fix(es) for index hopping issue. Currently its use is limited to sequencing large whole genomes where available UDI combinations can be utilized and not many are willing to use it knowing that their discoveries or positive expected results could be due index hopping.

Since index hopping seems to be caused by recombinase, recombining over a homologus sequencing/index primer binding sequence, what about using 454-MID style indexing - devote first 6-8 basepairs of the sequencing read to the index?

Than recombination events over flanking sequencing primers would not change much - the index would remain with its read.

Also, it would be great if Illumina could make a 4-channell version of the NovaSeq system/kits, since a lot of researchers would sacrifice half of the throughput for 3-10 times lower error rate. - Esp. if they want to assemble de novo umapped reads.

And one would be easilly able to run a 2 chanell sequencing kits on that system for NextSeq-style data.

I hope that the quality of the sequencing reagents keeps up with the time (Unlike the NextSeq V2 story): http://seqanswers.com/forums/showpos...0&postcount=89 and does not fall under BGISeq500 or similar system.

[nucacidhunter]

Originally Posted by Markiyan

Since index hopping seems to be caused by recombinase, recombining over a homologus sequencing/index primer binding sequence, what about using 454-MID style indexing - devote first 6-8 basepairs of the sequencing read to the index?

Than recombination events over flanking sequencing primers would not change much - the index would remain with its read.

The issue with inline barcode/index in Illumina systems is base diversity. In some library preps such as GBS which is done in batches of multiple samples it is easy but for a single library they have to have at least four colour balanced barcodes. Also if the inline barcodes were added to adapters then both R1 and R2 will read the barcode wasting a bit of sequencing.

[pmiguel]

Originally Posted by Brian Bushnell

I don't know anything about the lab issues, but the sequence quality is good. Coverage exhibits slightly more bias than HiSeq for the same libraries. Using unique dual barcodes and performing deduplication are both essential, because of the high crosstalk and duplicate rates.

Brian,
"HiSeq" doesn't denote a single kind of Illumina instrument. HiSeq 2000/2500 are quite different from HiSeq 3000/4000/X. Which type do you mean?

--
Phillip

[GenoMax]

I think @Brian is referring to 4- vs 2-color chemistry difference in that sentence.

[pmiguel]

Originally Posted by nucacidhunter

It will be good instrument when Illumina comes up with real fix(es) for index hopping issue. Currently its use is limited to sequencing large whole genomes where available UDI combinations can be utilized and not many are willing to use it knowing that their discoveries or positive expected results could be due index hopping.

Actually you can purchase full (96 or 384) UDI adapter sets from IDT. Although last I checked they were still a custom synthesis order. Which means you get more adapter than you are likely to use at a pretty high price. (Approaching $10K for the 96 UDI adapter set.)

Also, NuGen apparently has released a 96 UDI with some of their Illumina library construction kits.

--
Phillip

[pmiguel]

Originally Posted by GenoMax

I think @Brian is referring to 4- vs 2-color chemistry difference in that sentence.

Seems unlikely. That aspect was only introduced to the thread after Brian's post.

Brian actually praises the accuracy of the NovaSeq--which is what is claimed later is an issue.

I don't think that in actuality the two-color nature of the NovaSeq makes it less accurate than the HiSeq 3000/4000/X. I think it performs similarly or better. At least all Illumina's testing seems to show that.

--
Phillip

[Brian Bushnell]

Originally Posted by pmiguel

Brian,
"HiSeq" doesn't denote a single kind of Illumina instrument. HiSeq 2000/2500 are quite different from HiSeq 3000/4000/X. Which type do you mean?

--
Phillip

Oh, sorry, I meant OUR HiSeq machines Those are 2000/2500/1T. In this specific case I was comparing it to a 2500 run.

To clarify, from isolate random fragment data downsampled to the same number of reads in each case, after mapping, I observed slightly more genomic bases with very low (0x, 1x, etc.) coverage on the NovaSeq, and its primary genomic peak was shifted to slightly higher counts. We have not done a robust analysis of the effects on assembly, though I can't imagine coverage bias would give a better assembly. I will try to make some time to quantify this effect; it's of great interest to JGI as well. But we've only done 3 runs and I've only looked at 2 of them. Statistically it's not significant if you consider runs as the number of samples, but it is quite important if you consider reads as the number of samples. I'm not yet sure which is the case here.

[nucacidhunter]

Originally Posted by pmiguel

Actually you can purchase full (96 or 384) UDI adapter sets from IDT. Although last I checked they were still a custom synthesis order. Which means you get more adapter than you are likely to use at a pretty high price. (Approaching $10K for the 96 UDI adapter set.)

Also, NuGen apparently has released a 96 UDI with some of their Illumina library construction kits.

--
Phillip

Thanks. IDT has not released their UDI adapters that they are co-developing with Illumina. NuGEN adapters are not sold as a stand alone product and I am not sure if their adapters are specific to their kits or if they can be used with other kits as well.

[Brian Bushnell]

It's interesting to me that Illumina introduced NovaSeq without accompanying adapter kits to enable a high degree of multiplexing. Their current 24-unique-index kit seems targeted at human exome-capture on NovaSeq. From Illumina's perspective, higher degrees of multiplexing are probably seen as hurting profitability.

[nucacidhunter]

Originally Posted by Brian Bushnell

It's interesting to me that Illumina introduced NovaSeq without accompanying adapter kits to enable a high degree of multiplexing. Their current 24-unique-index kit seems targeted at human exome-capture on NovaSeq.

Illumina's current indexing strategy (up to 384 samples for Nextera) can be used on NovaSeq but using current TruSeq adapters only 8 unique combinations of index 1 and 2 is possible which is not enough for lots of applications.

[Markiyan]

Originally Posted by nucacidhunter

The issue with inline barcode/index in Illumina systems is base diversity. In some library preps such as GBS which is done in batches of multiple samples it is easy but for a single library they have to have at least four colour balanced barcodes. Also if the inline barcodes were added to adapters then both R1 and R2 will read the barcode wasting a bit of sequencing.

In that case we need to have first 4-5 bases as a random sequence (for clusters ID), followed by the barcode (another 4-8 bases), than the insert...

This is really important for RNAseq & CHIPseq applications, where we want to keep the cross-talk bellow 10^-6...
Classical dual unique indexing would get us into 10^-4 - 10^-5 range.

BTW: Actually the spatial signal separation from the patterned flowcell clusters should resemble more the 454 signal, and the clusters should be much more easier to call, even if low diversity at the first few bases.

[Brian Bushnell]

Originally Posted by nucacidhunter

Illumina's current indexing strategy (up to 384 samples for Nextera) can be used on NovaSeq but using current TruSeq adapters only 8 unique combinations of index 1 and 2 is possible which is not enough for lots of applications.

It only works for applications that are not sensitive to crosstalk. Personally, I would never multiplex samples of the same genus on a NovaSeq unless all libraries had dual unique barcodes. The current system of unique pairs, in which multiple libraries share the same first barcode and multiple libraries share the same second barcode, and any pair of first and second barcodes lands a read in a specific library, is insufficient for NovaSeq. To be honest, it was insufficient for the HiSeq 2500 as well, in situations where low-frequency variants were important (or for multiplexed single-cell sequencing). But it's more insufficient now, where multiplexing is desired on a NovaSeq, due to the substantially increased crosstalk rate.

[nucacidhunter]

I agree Brian. It will be even worse if they release S4 flow cells without individual lane loading device/fix which currently is in beta testing.

[nucacidhunter]

Originally Posted by Markiyan

In that case we need to have first 4-5 bases as a random sequence (for clusters ID), followed by the barcode (another 4-8 bases), than the insert...

This is really important for RNAseq & CHIPseq applications, where we want to keep the cross-talk bellow 10^-6...
Classical dual unique indexing would get us into 10^-4 - 10^-5 range.

BTW: Actually the spatial signal separation from the patterned flowcell clusters should resemble more the 454 signal, and the clusters should be much more easier to call, even if low diversity at the first few bases.

Adding 4-5 diversity base is possible but two points to consider:
1- it will be difficult to obtain good oligo annealing to form adapters and also there will unpaired oligos which would require further purification
2- Illumina is unlikely to change their established indexing strategy and also there might be some patents on inline barcoding

Unique duel indexing (every library with different index 1 and 2) should enable identification and filtering of any fragments resulting from index hopping.

Actually in patterned flow cells position of each nano-well which would have clusters is already known. I have not thought about the need for sequence diversity but it could be related to background signal.

[pmiguel]

Originally Posted by nucacidhunter

Thanks. IDT has not released their UDI adapters that they are co-developing with Illumina. NuGEN adapters are not sold as a stand alone product and I am not sure if their adapters are specific to their kits or if they can be used with other kits as well.

You can buy the UDI 96 and/or the UDI 384 kit from IDT as a custom order. Just contact their sales. We bought the UDI 96. Made libraries (both no Amp DNA and RNAseq libraries) from them. Ran them on our NovaSeq.

The UDI 384 are 10 base indexes though. So you would have to reduce your sequence insert reads by 2 bases each to compensate...

NuGEN sales reps told me that their adapters don't use T/A tailing. So they wouldn't work with most other kits.

--
Phillip

[pmiguel]

Originally Posted by Brian Bushnell

It only works for applications that are not sensitive to crosstalk. Personally, I would never multiplex samples of the same genus on a NovaSeq unless all libraries had dual unique barcodes.

Did you see much phiX index hopping? Searching the "undetermined" fastq for index hops involving one of i7/i5 GGGGGGGG/TCGTAGTG I tentatively identified a fair number. Of course they may just be the result of signal bleed.

--
Phillip