454 Sequencers for Eukaryotic Genomes

[snape_ar]

I would like to know the reason why we don't use 454 Sequencers for for sequencing Eukaryotic Genomes (ChIP-Seq, DNA Seq, RNA-Seq) and why most of the time people use Illumina ?

Is it because of the error rate in Illumina platforms is low as compared to 454 although the read length is 4-5 times more in 454 ? Or, there are some other reasons ?

[pmiguel]

It is because the cost/base of sequence using the 454 is on the order of 100x that of an Illumina HiSeq2000.

The 454 gives you much longer reads than are possible on an Illumina HiSeq (~500 bases for a GS-FLX and around 700 bases for a GS-FLX+ vs 100 bases on the HiSeq.) But the Illumina allows "paired end" reads from each DNA fragment, so it is doing 2x100 bases. 454 run times are short -- about a day, whereas a 2x100 run on the HiSeq takes nearly 2 weeks.

So, at 10x the cost/base, I think a 454 would compete on the basis of read length. But at 100x it is, well, pretty much obsolete.

--
Phillip

[jimmybee]

It all depends on the project you have going. Yes, Illumina is cheaper and gives much more data, but a lot of people rely on the larger read length when dealing with complex genomes that contain large repetitive elements. The development of longer read length in Ion torrent, and PacBio technology add significant value to projects for this exact reason. Its all about using a technology to suit your sequencing needs.

[pmiguel]

Sorry jimmybee,
Not buying it. What you write sounds completely reasonable, but ignores what the original poster asked and the larger issue of "how much is a longer read worth." I hazard "10x" more above for substantially longer reads like those offered by the GS-FLX+. I am willing to consider numbers somewhat above and below that figure. But at 100X more, I don't see a rational argument to be made.

PacBio and Ion Torrent/Proton are both different stories. The former is likely DOA with the release of Oxford Nano Pore Technology's Min/Gridion sequencers (later this year?) The latter, especially the Ion Proton, may actually challenge Illumina. Or not.

I think the truth is that as a rapidly developing platform some years ago, the 454 was competitive with Illumina. But their recent actions remind me of Beckman's back in the early capillary sequencer days -- they disregarded their respective leaders in the markets where they competed and were inevitably crushed in those arenas.

Just like Beckman a decade ago, Roche Applied Sciences now pursues a nearly solipsistic marketing and development path. Of course they are free to do as they like, but I'm not going to pretend like their outlook bears an even vague relationship to the real world.

--
Phillip

[jimmybee]

I definitely don't disagree with your comments right there. And a lot of what I mentioned is dependent on resources. You're right, at the prices per base there is no way anyone should really look at doing GSFLX over Illumina. Its just not competitive to Illumina.

What I was merely pointing out is that some projects rely on 454 compared to Illumina and have little options to get around it. We've had massive problems looking at wheat and barley with only Illumina, so we've had to use 454. Thats why the development of read length in Ion torrent looks really really promising. Its not going to replace Illumina's cost-effectiveness, but will help guys like us that need that read-length to overcome specific problems.

[TonyBrooks]

I definitely don't disagree with your comments right there. And a lot of what I mentioned is dependent on resources. You're right, at the prices per base there is no way anyone should really look at doing GSFLX over Illumina. Its just not competitive to Illumina.

What I was merely pointing out is that some projects rely on 454 compared to Illumina and have little options to get around it. We've had massive problems looking at wheat and barley with only Illumina, so we've had to use 454. Thats why the development of read length in Ion torrent looks really really promising. Its not going to replace Illumina's cost-effectiveness, but will help guys like us that need that read-length to overcome specific problems.

The main advantage of Illumina over 454 is read number. You typically need 10s of millions of reads to do RNA-Seq, ChIP-seq etc. Typical output of our 454 is around 1.5 million reads (shotgun).

Also on the horizon is the The MiSeq Q3/4 upgrade which promises 10m million reads of 2x250bp for 1/10 the cost of a 454 run. With some stringent library prep, there's no reason why you can't generate 450bp reads on MiSeq (using overlapping paired end reads). That's pretty comparable to the FLX Titanium lengths we're seeing on our 454. Sequence quality should also in theory be better. Q-scores decrease with read length, but the final 50 bases of a 250bp paired end 450bp library will have been read twice, reducing error rate.

[pmiguel]

I definitely don't disagree with your comments right there. And a lot of what I mentioned is dependent on resources. You're right, at the prices per base there is no way anyone should really look at doing GSFLX over Illumina. Its just not competitive to Illumina.

What I was merely pointing out is that some projects rely on 454 compared to Illumina and have little options to get around it. We've had massive problems looking at wheat and barley with only Illumina, so we've had to use 454. Thats why the development of read length in Ion torrent looks really really promising. Its not going to replace Illumina's cost-effectiveness, but will help guys like us that need that read-length to overcome specific problems.

What is the wheat or barley genome problem that is intractable with Illumina sequence read lengths but is solved using 454 read lengths?

I am familiar with wheat and barley genome structure. Yes, a nightmare that Illumina is not going to be able to traverse de novo. But I still don't see how increasing your read length 10x helps much.

Unlike animal genomes, plants tend to have compact genes, with introns only rarely longer than a kb or two. So you could sequence genes de novo and attempt to string the genes together in their correct order/orientation using mate ends. Won't work in many instances because even a moderately-sized retro cluster will create a repetitive block large enough that even 20 kb mate ends will not be able to bridge[1].

Point being whether you have 100 bp or 700 bp reads, you still have the same basic limitations retro-cluster-wise. So you might as well go with the much lower cost and higher accuracy of Illumina sequence.

Small (non-autonomous) transposable elements may not be traverse-able in a single read at 100 bp, but with Illumina paired-end reads to anchor your middle repetitive TE read with another low copy read in the surrounding sequence, you are back in the game with a little gap filling informatics. And, even without that, you should be able to span these small elements with mate end reads to create scaffolds.

So the question remains, what are 700 bp reads giving you that makes them worth their 100x cost vs paired 100 bp reads?

--
Phillip


[1]Well, probably. I could envision a method using pretty high coverage 20 kb mate ends. So maybe using a cosmid-based mate end library construction method, you could do it. The idea would be to first determine the LTR structure of all high-copy LTR retrotranspons in the genome, then find the unique junctions created by the insertion of elements into one another. These junctions then serve as the "stepping stones" across the retro cluster swamp.