My notes from Helicos' UBS presentation

[ECO]

These are rough...and I can't find some of them on the Helicos web. Just thought I'd throw them up here if you don't want to go listen for yourself for 23 minutes.

50MB / hour out the door, 10-fold increase in 2 years to get to 500mb per day (>???) no addl capex for upgrades in perf.
10GB/run, 8 days
400 million reads (25-50 bp)
50 channels

<5% error rate, 0.5% substitution rate
error rate and length of read are independent
no GC bias (no sample prep...high call rate, no skewing (??)...basically due to no amp)

chemistry performance independent of template size (25bp - 8kb).

paired end read, sample prep free (on instrument)

per sample costs
$360 kit -->$381 w/external req'd reagents = 8 million reads (one lane)

EA instrument progressing well
2nd from major US cancer center, ship ~Oct 1st
13 in assembly phase, 8 can run samples
15-30 by mid 09

-----
business

10-fold increase in 2 years to get to 500mb per day (>???) no addl cap exp

single use kits
sequencer
linux cluster

$1.35 million list
kits $18,000 ($600k/year instrumnet, 900k within first year) (my calc: ~33 kits year = 10 day runs? = <1 GB/day).

$5M revenue stream 60-70% margin per instrument

Scalable production very hard (workflow), complex sample prep
tSMS: amp free solution

comparisons to leading NGS platform:

10Gb/run vs. 4.5Gb per run GAII as of 9/25 (didn't mention 8-10d vs. 3-4d)

100Gb/run with no hardware improvements

digital gene expression
candidate region resequencing
50 samples per run (vs 8)

6% sample premium (reagents beyond what's provided from HLCS) (265% with others), this is mostly all the library prep reagents. Not counting sample prep labor.

Milos presented "ton of data" on DGE and CRR in providence

[genseq]

Dinosaur!
It must die out!

[Chipper]

?????????????

[genseq]

"... 10-fold increase in 2 years to get to 500mb per day ..."
"<5% error rate ..."

[Chipper]

Ah, ok it should have been 500 Mb / h then since the current spec says > 1Gb per day. But you are right, Pac. bio may make all other instruments look like dinosaurs by then, who knows.

[genseq]

Sorry for this mistake.

But how about "<5% error rate ..."?

[Chipper]

Well, it is less than 5%... But we will see when they present new data if it is a problem or not. The high rate is due to the single molecule sequencing, if a nucleotide is "dark" it will look like a deletion since it will not be compensated by other strands in a cluster like the Solexa chemistry, but it should be possible to take this into account when aligning to a reference. On the other hand you do not get the high error rates at the end of sequences since there is no risk that the syntesized strands will be out of phase when it is only one single strand...

[genseq]

Too many problems:
1. 5% looks like a deletions;
2. 1 350 000 $ ;
3. $600k/year instrumnet, 900k within first year ;
3. 2000 kg .
4. Very short reads ;
5. Photoinactivation of DNA (limits prospects);
6. et cetera

[Chipper]

I am interrested in your fifth point, what does it mean? You forgot this point also: altitude < 1500 m... (from spec. sheet). But to clarify, 1. can be handled in alignment or by multipass sequencing / consensus base calling. Hard to say anything about the price before it delivers data, it depends on where the competitors are and how much you gain by tSMS.

[genseq]

Originally Posted by Chipper

I am interrested in your fifth point, what does it mean? You forgot this point also: altitude < 1500 m... (from spec. sheet). But to clarify, 1. can be handled in alignment or by multipass sequencing / consensus base calling. Hard to say anything about the price before it delivers data, it depends on where the competitors are and how much you gain by tSMS.

1. "altitude < 1500 m... "
- depends from damage of CCD by ultra-rays.

2. "5. Photoinactivation of DNA"
- depends from photodamage of the nearest to fluorophore nucleotides by singlet oxigen and other excited molecules. Result - short reads with bad prospects.

3. "10-fold increase in 2 years to get to 500mb per hour"
- and how about competitors in 2 years?

[polytoo]

Originally Posted by Chipper

Ah, ok it should have been 500 Mb / h then since the current spec says > 1Gb per day. But you are right, Pac. bio may make all other instruments look like dinosaurs by then, who knows.

Hopefully that is true. But until PacBio ships a box the Helicos is the only single molecule sequencer. Anyone sick of troubleshooting the building of libraries? Of looking at skewed data because of amplification bias?

[Chipper]

Yes, I am. Maybe I should make another try to get some reads . You don't happen to know where to find an instrument?

[polytoo]

Across the hallway, yup, there is one in there.

[Nitrogen-DNE-sulfer]

I tend to worry more about the esoteric enzymes used in next gen which are still in the 1980s in terms of their understanding or publication record and Taq is not one of them.

PCR is a far more understood bias than people appreciate. Note how complete genomics amplifies the whole genome in 400-500bp fragments and still gets over 90% of the genome with short noncontiguous reads (also not 1% DMSO and Glycerol but still quite amazing). Far more than the 87% Quake published on Helicos. Sorry his published 90% is a fraud unless he's expecting to be missing a few chromosomes in his own genome...2.7Gb as their genome size denominator helps to inflate the 90% covered number published. Reviewers should be ashamed of this 158Mb (size of some model organisms) oversight or at least justify which chromosome they choose to ignore to reduce the genome from 2.858Gb to 2.7Gb).

Back to the only niche statement they have.. Bias free.
Have a look at the enzyme Helicos uses to put their PolyAs on. Only a handful of papers exist (vs 1000s on PCR) and they all scream extreme bias with secondary structure quite prevalent in RNA. No PolyAs... no templates to sequence. This is a very binary bias. Bulk library PCR can be eliminated for 454, ILMN, and SOLiD, as shown in Kozarewa et al and I'd much rather emulsion PCR or Bridge PCR than bulk Poly A a library as the former are non-darwinian reactions (each molecule not competing with the other as they are geometrically isolated..perhaps more so for emPCR than clusters). Likewise, I dont see how one can say high error and low bias in the same sentence. Seems oxymoronic unless one truly believes all errors are non-systematic.


Characterization of the E.coli poly(A)polymerase: nucleotide specificity, RNA-binding affinities and RNA structure dependence

Shlomit Yehudai-Resheff and Gadi Schuster*
Department of Biology, Technion√Israel Institute of Technology, Haifa 32000, Israel

Polyadenylation of RNA molecules in bacteria and chloroplasts has been implicated as part of the RNA degradation pathway. The polyadenylation reaction is performed in Escherichia coli mainly by the enzyme poly(A)polymerase I (PAP I). In order to understand the molecular mechanism of RNA poly adenylation in bacteria, we characterized the biochemical properties of this reaction in vitro using the purified enzyme. Unlike the PAP from yeast nucleus, which is specific for ATP, E.coli PAP I can use all four nucleotide triphosphates as substrates for addition of long ribohomopolymers to RNA. PAP I displays a high binding activity to poly(U), poly(C) and poly(A) ribohomopolymers, but not to poly(G). The 3'-ends of most of the mRNA molecules in bacteria are characterized by a stem√loop structure. We show here that in vitro PAP I activity is inhibited by a stem√loop structure. A tail of two to six nucleo tides located 3' to the stem√loop structure is sufficient to overcome this inhibition. These results suggest that the stem√loop structure located in most of the mRNA 3'-ends may function as an inhibitor of poly adenylation and degradation of thecorresponding RNA molecule. However, RNA 3'-ends produced by endonucleolytic cleavage by RNase E in single-strand regions of mRNA molecules may serve as efficient substrates for polyadenylation that direct these molecules for rapid exonucleolytic degradation.

[maggie]

Please delete this post. Thanks.

[polytoo]

Originally Posted by Nitrogen-DNE-sulfer

I tend to worry more about the esoteric enzymes used in next gen which are still in the 1980s in terms of their understanding or publication record and Taq is not one of them.

One could worry about the esoteric until the cows come home but the proof is in the pudding. Taq polymerase has GC bias among other biases. When starting with very small amounts of starting material or having to do many cycles the experiment that requires quantitation is going to get muddy.

A quote recently from Pater Park pasted from a Genome Web interview:

Specifically, the authors noted that in the Illumina data, they observed a "modest over-representation" of reads from regions of G-C content of around 40 percent to 65 percent, which they attributed to "bias introduced by PCR or cluster amplification." The Helicos data, meantime "had a relatively even distribution across 20-80 percent G+C content."

In addition, the researchers were able to use much less starting DNA with the Helicos system than with the Illumina — 50 picograms as opposed to 5 nanograms.

We have found the same thing. And without any amplification the ChIP-Seq experiment is just a whole lot easier from start to finish.

[Nitrogen-DNE-sulfer]

I'd caution that paper in particular.
Look at the GC bias they show on the ILMN. That's a pretty sloppy ILMN run and if you look at the supplement you can see they amplified the library AND used cluster PCR so I dont see how they can come to the conclusion that its cluster PCR induced in light of Kozarewa et al. I'm making the assumption they used the 18cycle default ILMN amp protocol which is well know to generate this so I dont see the point of the paper. Its a bit obvious from Kozarewa et al that this is what one expects from a competitive amplification....hence my earlier point about PAP I and the fact that this reaction is competitive and not geometrically constrained like the cluster PCR suggested in that Genomeweb article. I think pudding statements are fair and so far I see CGI with more than 90% of the genome (and they amplify more than anyone) and Helicos with less than that. For that matter I think Bentley et al is better than 87% on the YRI human, likewise SOLiD and potentially even 454 at 7X with watson (although too expensive). So I'm not quite convinced on that work that there is something there yet.

Pairs are critical for survival and until they show that this isnt impossible due to photo-damage or other issues, my attention is in other places.

On the note of pudding...Helicos is far more impressive than PacBio...hands down. Must be their glamorous fireworks or marketing but nothing adds up on that platform..25,000 1Kb reads for $100.... Assuming minimal re-read rates (4x) this is north of $10,000/Gb and 454 like hundreds of Mb/day. Someone need to emphasize that things are heading to $/Gb not $/Mb very soon and they are not the saving grace here. I see them niche' players for hospital like turn around times for infectious disease (where price doesnt matter and time does) but just dont see a pure Human play until CCDs and optics get 5,000 times cheaper than XY stages.
They will supplement without a doubt for phasing but the moment a ILMN/SOLiD/Helicos can phase with a library trick the economic difference is so vast that I suspect that strobed feature wont matter anymore.

[polytoo]

You bring up the lower coverage for complete human resequencing of Helicos and I might agree. But it is not the right tool for the job. Let Complete Genomics do the boring job of resequencing human. ChIP-Seq, RNA-Seq, GRO-Seq, BS-Seq, DGE, is where the action is. It is the number of unique hits in ChIP-Seq and the other protocols. Helicos seems to kick butt in that arena. By any true measure the dollar per unique hit is more than a tad better than anything else. If you factor in the complete pipeline with FTE it is quite a bit cheaper. Not having to factor out amplification is good. Not having to amplify at all makes the whole thing fun.

As for the resequencing human we are doing a human resequencing but only as a back burner project just to see how it would go. With just a few runs I am very impressed with the data. But thankfully the big genomic places will do that tedious task. Or we send our $5,000 to Complete Genomics. Not competing with that and I love them for doing it.

Helicos is far more impressive than PacBio

I would agree but would also add that it is also comparing a working instrument to words. But even as you say it works as described they are creating a machine that has a niche I have a hard time conjuring up.