Neil Hall is Responsible for Rising NGS Prices

“Wait, what?! Neil Hall is responsible for NGS prices actually inching up rather than continuing down to ever lower depths?”

No, not really.

Well, he kind of is. And most likely so is everyone reading this blog. (All three of you.)

It’s because DNA sequencing consumers didn’t ramp up their usage as quickly as Illumina was driving down the prices. When Illumina bumped HiSeq capacity from 300Gb per run to 600Gb per run, their customers said:

“Yay! The same amount of sequencing for half the price.”

To which Illumina responded:

 “What? No, no, no, no! You’re supposed to sequence twice as much. Actually, you’re supposed to sequence even more because the cost is so low that new applications are now feasible”.

And the stock market said:

“Hmm, I’ll have half that market cap back, thank you.”

So Illumina learned their lesson and decided to be more careful about releasing upgrades for the HiSeq. Right around the time they launched the 600Gb system, they said they were able to generate runs of over 1Tb in house. It’s been around two years since then, but still no update.

Why not? Because there isn’t any market pressure to do so. The only competition HiSeq had was SOLiD, but that platform has essentially been abandoned by Life Technologies as a commercial failure. (You can love SOLiD reads all you like, but even with a much bigger marketing and sales budget, it was still getting trounced by HiSeq. Too bad they didn’t release Wildfire a couple of years earlier – we might not be having this conversation right now.)

And while Illumina was busy dominating the high throughput NGS market, Ion Torrent came out and tried to carve out a different niche: much faster, much cheaper machines (albeit with lower throughput). And that niche has really taken off (especially in the clinical space). Illumina responded with their own machine (the MiSeq) and since this is the only space where they’re facing serious competition, it’s where they are concentrating most of their resources. This means we’re now getting much faster reads, but they aren’t really getting any cheaper.

So is the party over for cheaper reads? Since the ‘plateau’ was caused by market forces and not technological advancements, the answer is almost certainly “no”. Researchers will figure out ways to creatively use ever more sequencing capacity, and when they start bumping into the limits of what the current machines can do, those machines will almost certainly improve. And the competition could come back. Ion Torrent has stumbled on their original road map, but if they ever release the PIII chip, that could put pressure on the HiSeq. And while Oxford Nanopore is the latest to suffer from the hype-hate cycle, there are a lot more companies waiting in the wings to dethrone Illumina. Check out BlueSEQ’s ‘Emerging Technologies’ section of the NGS Knowledge Bank for a partial list.

Dr. Hall is right that the field would benefit from genome scientists focusing on “new ideas”.  He’s also right that the cost of sequencing will plateau. But I don’t think we’re quite there yet. And where it plateaus, whether that’s $1000, $100 or $10, will make a big difference. 

Edit: Gah! I totally forgot to include a link to Mick Watson's excellent take on the cost of sequencing.

X-Gen 2013

I recently had the pleasure of attending X-Gen (more formally known as ‘X-Gen Congress & Expo). This meeting falls on the small side of the spectrum, with a couple hundred attendees, and that suits me just fine. I prefer smaller meetings as I find it easier to network and generally keep track of what’s going on. It does, however, have parallel tracks – RNA Sequencing, DNA Sequencing and Data Analysis – so hard choices sometimes had to be made (e.g., a talk on Illumina’s Moleculo technology or one on duplex sequencing for rare mutation detection; Moleculo won out, but I really wanted to be at both). I’m not quick enough for live blogging – I can type up notes/thoughts, or I can listen, but I can’t do both. Therefore, I’ll just mention some of the things that stuck out in my mind.

• From Peter Park’s talk on the role of retrotransposition in cancer – 45% of the human genome is derived from transposable elements (most of which are under epigenetic suppression). This isn’t a new discovery, but I still find it fascinating. Same way I find it fascinating every time I hear about how our bodies harbor ~100 trillion microbial cells (dwarfing our ‘own’ 10 trillion cells). Anyway, this analysis is fairly tricky (presumably due to the difficulty of sequencing/aligning repeats), but longer reads on the horizon, like that of the Moleculo technology, should help.
• John Stam discussed the role of gene regulation in common disease. He mentioned something that, while it intuitively makes sense, I still found really interesting. When he piled up the results from 50 separate transcription factor ChIP-Seq experiments, it matched the results from his DNaseI hypersensitive site ChIP-Seq almost perfectly. Genomic regions available for TF binding would need to be accessible and, therefore, hypersensitive to DNaseI, but it’s still kind of cool to see the results overlap so well. The other interesting thing he showed was that when he had very deep sequencing in a TF ChIP-Seq experiment, the resolution of the binding sites became so good that it matched with the predicted sites of the TF crystal structure. Seeing two completely orthogonal approaches match so closely is pretty amazing.
• David Smith talked about studying head and neck cancers, first with RNA-Seq and more recently looking at very large genes and chromosome breaks. For example, the FHIT gene at 1.5 million bases is so large that transcription takes so long that it can interfere with replication (hence the breaks). While it isn’t directly related to the science he was discussing, Dr. Smith revealed an interesting marketing tactic (one of the worse kept secrets in the business) about Life Technologies: “SOLiD gave us $300k worth of free sequencing… back in the days with sequencing companies were willing to seduce you.” Very important to keep in mind when companies throw around market share numbers.
• Cynthia Burrows is using an alpha hemolysin-based nanopore system from Electronic BioSciences (EB) to look at DNA damage and other modifications. EB is one of the nanopore companies flying under the radar. They use a small aperture capillary with a single nanopore and a dual electronic system so that translocation and measurements can be kept separate. Check out the EB entry in BlueSEQ’s ‘Emerging Technologies’ section. Her fun fact for the day is that there are ~10k spontaneous depurinations per genome in a cell per day. That definitely makes me feel like I’m falling apart!
• Aleksei Akimentiev gave a really interesting talk on solid state nanopores. While he didn’t have any experimental data to show, he showed some fantastic model simulations run on the ‘Blue Waters’ super computer. He has modeled all sorts of nanopores, including graphene layers which turn out to be quite ‘sticky’. He’s working with both Oxford Nanopore [http://blueseq.com/knowledgebank/emerging-technologies/oxford-nanopore-technologies/] and IBM [http://blueseq.com/knowledgebank/emerging-technologies/ibmroche454/]. Check out some of his animations here:
• silicon: http://www.nsf.gov/discoveries/disc_videos.jsp?cntn_id=117318&media_id=67049&org=NSF
• graphene: http://pubs.acs.org/doi/suppl/10.1021/nl301655d
• nanopore capacitor: http://pubs.acs.org/doi/suppl/10.1021/nl071890k/suppl_file/nl071890ksi20070802_123247.mpg
• Jingyue Ju is working with tagged nucleotides (with various PEG molecules acting as the tags, if I remember correctly). The DNA strand to be sequenced is processed by a polymerase which releases the tags as the nucleotides are incorporated. These tags are what are read by the nanopore. Genia announced last fall that they had switched their chemistry to use Dr. Ju’s system. When asked how long until he could sequence DNA, Dr. Ju answered “In a couple of months we’ll demonstrate real sequencing data”. Good news for Genia?
• Massimiliano di Ventra talked about “Fast DNA Sequencing via Tunneling” – not exactly a nanopore, but definitely related. It looks a really cool technology, but unfortunately all he has to show at the moment is simulated data. Perhaps the most interesting thing is that he brought a new commercial sequencing company to my attention – he’s been working with Osaka-based Quantum Biosystems (a new one we’ll need to add to the ‘Emerging Technologies’ section of BlueSEQ’s Knowledge Bank.

NGS Year in Review Part 1: Promises, Promises

In an annual ritual, the various NGS vendors gathered in Marco Island at AGBT to sell their wares and tout their technology to the eager attendees. I figured this makes it a good time to review what's happened over the past year. I’ll start by covering which promises were kept and which were broken, and follow up with subsequent posts on a more general overview of the progress we’ve seen, another one on the mergers that did and didn’t happen, and finish up with a post where I take a stab at what we'll see over the coming year.

2012 was a year of hope and promises. The NGS market is brutally competitive, with all the vendors trying to outdo each other, and 2012 was no exception. The major cycle of announcements that gives us a hint at what we were to see in the upcoming year starts with the JPMorgan Healthcare Conference and finishes with AGBT on Marco Island. 

Last year Ion Torrent (Life Technologies) kicked everything off by announcing their new machine, the Proton. While many, myself included, groused that this new machine made a mockery of their slogan 'The chip is the machine', it was necessary for them to keep pace with their previous promise of improving outputs 10-fold every 6 months. The first Proton chip, the PI, was slated for mid-2012 and would generate 10Gb/run. Not bad, but it was the following PII chip, slated for the end of 2012, that caught everyone's attention. If it were to keep up the trend of 10-fold improvements, it would generate 100Gb of data in a few hours. Better yet, it would finally achieve the holy grail of NGS - the $1000 genome! But Ion Torrent didn’t quite pull it off. Despite making progress, their schedule has started to slip and the output specs are starting to look more meager. Combined, it means they didn’t hit the fabled $1000 genome in a day. Not even close. By the end of 2012 (and through AGBT 2013), their most advanced commercially available chip, the PI, generates around 10Gb of output per run. At $1000 per chip, it would cost around $9k and take up to a week to sequence a human genome. Getting better, but still not there. The chip they said would reach the goal, the PII, was first delayed to March 2013 and now to ‘mid-2013’. Also, rather than generating 100Gb at launch, they’re now talking about 32Gb/run at launch, climbing to 64Gb over time (probably about 12 months). So we might not see the “$1000 genome in a day” from Ion Torrent until mid-2014, or even a bit later.

Illumina got much closer to fulfilling their promises, but that’s partially because they weren’t so bold. While they did claim they’d have the ‘genome in a day’ by the end of 2012, they never said anything about what it would cost. The machine to perform this trick, the HiSeq 2500, was released and the official specs at the end of 2012 were pretty close – 27 hours for a 30X genome (90Gb). With recent improvements (especially on sample prep and analysis), they should be able to go from DNA to analyzed sequence in under 24 hours.

Ion Torrent and Illumina made their big announcements at the JP Morgan Healthcare conference, but Oxford Nanopore (ONT) saved their news for the more scientifically oriented AGBT meeting. And by all accounts, their short presentation stole the show and set Twitter afire. While they had already given us a peak at their upcoming GridION system, they started putting some specific numbers behind it, with the most dramatic being a 20-node system that by 2013 would sequence a genome in 15 min for ~$1000 ($10/Gb). Oh, and they gave a peak at some really intriguing data – 50kb reads with a 4% error rate (that would drop to 1% by the launch). But then they really stirred things up when they unveiled the MinION, a disposable handheld nanopore sequencer the size of a large USB stick, suitable for field work (including no need for sample prep), that would cost under $900 and generate over 1Gb of data. And both the GridION and MinION would launch in 2012.  Nanopores were finally here! These claims were met with incredible excitement and even appeared to shake up the emerging NGS space by, allegedly, prompting Halcyon Molecular to close up shop.

Almost immediately, however, the cries of “Show me the data!” started to emerge from the ranks of scientists. As mid 2012 approached with no new data, the shouts became stronger. Some, myself included, held out hope when ONT was scheduled to appear at ASHG. A booth with pretty but non-functional demo units was all we got – still no data. When the agenda for AGBT 2013 came out, ONT was nowhere to be seen, so any hope of a last minute nanopore miracle vanished. But ONT did attend the show, and CTO Clive Brown gave an impromptu interview to noted blogger Nick Loman that seemed to reveal some of the issues that they were running into and a maybe/sorta indication of a 2013 launch (or at least data from early access customers). While many feel burned by ONT's undelivered promises, the truth is that all will be forgiven once they get their hands on a functioning nanopore sequencer. Book your tickets now for AGBT 2014. 

[If you’d rather listen to me speak than read my words, Theral Timpson interviewed me about this over at Mendelspod.com]

 

2012 Will NOT be The Year of the $1000 Genome

There was a lot of excitement earlier this year when Ion Torrent announced their newest machine, the Proton. It was going to let them continue down the path of rapid technology improvement and let them keep their oft-repeated statement of 10-fold improvement every 6 months. I was pretty skeptical of these claims when they first came out, but the PGM has done a remarkable job of backing them up, going from 10Mb on the 314 chip, to 100Mb on the 316 chip and finally to 1Gb on the 318 chip over a period of 12 months.

As impressive as Ion Torrent’s progress has been, there have been a few hiccups:

• Contrary to their initial marketing slogan, the chip is no longer the machine. To keep riding the wave of improvements, you have to trade in your original PGM for a much more expensive (but still relatively cheap) Proton. Not surprisingly, there were some grumblings from customers when that little bombshell came out.
• They have slipped a bit (by about 3 months) on their rapid pace of launching a new chip every 6 months. By their original plan, the PI should have come out in July, but we’re just now seeing it in September.
• When the Proton was first announced, the PI was targeted for 10Gb, or 2 exomes per chip, and the PII was to offer 10 times the output which would be 100Gb, enough fo a single whole human genome (all for under $1000 for chip+reagents). The only problem is, they’re now redefining “whole human genome” to be 20X coverage, while the industry standard is more like 30X with some pushing for higher coverage of 40X or more. 20X coverage implies that the PII chip will initially only be generating 60Gb per run. Still quite impressive, but another slip from the original roadmap.

So, this means not only will we not have a whole genome on a single chip by the end of the year (due to the 3 month schedule slip), when we do get it, it won’t really be a ‘true’ whole genome run. You’ll probably need to run two chips (or a chip and a half, if you can manage that with indexing), meaning it will cost you more like $2k instead of the holy grail of $1k.

The other thing to note is that the launch of the PII will signal the end of the original ’10-fold improvement every 6 months’ promise. The follow-on chip, the PIII, will take a full 12 months after the launch of the PII, and it will only be about a 2X improvement instead of the 10X improvement we’ve been getting used to.  Still, if they’re able to do what they say, by the middle of 2014 we’ll have the ability to generate ~200Gb in a 2-4 hour run for ~$1000. That, in turn, will surely put a lot of pressure on the market leader, Illumina, to launch some improvements of their own. Chemistry A and B, we’re waiting… 

NGS Hype War

Who’s winning the NGS hype war? 

A few years ago it was PacBio – they were going to sequence a genome in 15 minutes. Currently it would take them >90 days and cost ~$225k. (They’re now more sensibly focusing on the advantages of long reads and directly reading modified nucleotides.)  

Ion Torrent took over the reigns of ‘hype king’ with the launch of the Personal Genome Machine.  They are building on the trillion dollar investments of the silicon wafer industry and promising 10x improvements every six months, to infinity and beyond! The most recent promise is a $1000 genome in two hours in 2012. In all fairness, they’ve kept up their promises over the past 12 months. The “10MB” 314 chip is now producing up to 40MB. They launched the 10x improved “100MB” 316 chip over the summer, and it’s now routinely generating >200MB of data. The “1GB” 318 chip hasn’t launched yet, but it seems reasonable that they’ll be able to hit their target. I haven’t heard anything about their next chip (what would/could be the 320 chip), but I suspect they’ll have lots to say at Marco Island.

So Ion Torrent is clearly currently winning the hype war, but is there a contender? Mabye.  

Genia is a new start-up out of Menlo Park. They’re talking about the $100 genome (the $1000 genome is so 2011) and they’re calling themselves the “Last-Gen Sequencing” company. Apparently their technology is so revolutionary there won’t be a need for anything else. Just a heads up to anyone else developing their own technology, you might as well stop. Genia’s got it covered. No, they don’t have any instruments to sell just yet. And, no, they technically haven’t actually produced any sequence yet. But just you wait. They hype has just begun!

What's the NGS Buzz?

There's been quite a bit of 'next-gen' sequencing news lately (some of it positive, some of it pretty negative). I thought I'd take a stab at ranking them in terms of their "buzz" - how much people are talking about them, excitement around updates, etc. It is very subjective and definitely NOT an attempt at ranking their usefulness/performance (which would be highly application specific). Feel free to let me know how wrong I am in the comments section.

(I inexplicably left out Complete Genomics - now corrected)


 

Platform	Comments	Buzz Factor	Trend
Ion Torrent	Lots of promises, excitement and chatter Lots of blogs analyzing data	10
MiSeq	Data starting to trickle out Mostly in a battle with Ion Torrent	6
GnuBio	Completely backed off their "$30 WGS" claims Interest picking back up now that they're in beta testing	5
Complete Genomics	Still in the news, in a death battle with ILMN and BGI	5
HiSeq	Dominant (if boring) NGS workhorse	4
Oxford Nanopore	Not much news Initial excitement waning Waiting for next (first?) big announcement	3
Pacific Biosciences	Only news is bad news Management overestimated market adoption of PacBio Laid off 28% of workforce (ouch!)	3
SOLiD	Has become the ugly stepsister of Ion Torrent Management says "putting all chips into Ion Torrent" and "(light based seq) is nearing the end"	2
454	Not a lot of news since FLX+ Not much in the way of public discussion	2

Helicos Cries Out “I'm Not Dead! I'm Getting Better!”

Helicos, the first company to launch a single molecule DNA sequencer (a class sometimes called “3^rd Generation”), has been hanging on to a thread for a while. They've been delisted from NASDAQ and their cash supply is dwindling (down to $1.6M as by June 30^th). They are going through another round of cash-saving layoffs that will bring their headcount down to 10. They have apparently stopped selling the HeliScope and their ability to supply reagents and service support to their current customers has been called into question.

When they launched their HeliScope in 2009, they went head to head with the more established players (Illumina, ABI and 454) focusing almost exclusively on being the first 'true single molecule sequencer' for genomic sequencing. It was a tough sell given that they were a bit late to the party and that their reads are shorter and have a higher error rate. They might have been better served to focus on RNA-Seq and ChIP-Seq where the lower quality reads wouldn't have hurt as much and they could have touted their (at the time) industry-leading # of reads per run.

For the past few months their main strategy seemed to center around litigation – they're suing Life Technologies, Illumina and Pacific Biosciences for patent infringement. Just recently, however, they've expanded this strategy by borrowing a page from Complete Genomics – they have launched a sequencing services program to increase their revenues. They've also launched an interesting targeted sequencing solution in which the capture sequences are attached directly to the flow cell (rather than their standard poly dT sequences). They've demonstrated this technology by creating a flow cell for the direct sequencing of the BRCA1 gene.

I fear it may be a bit too little too late, but at least Helicos is still kicking and demonstrating that it doesn't want to go on the cart.

Ion Torrent's Roadmap to the $500 Genome

Ion Torrent has made some bold claims about making 10x improvements every 6 months, and so far they've kept to that schedule (albeit only over a single 6 month period). In a recent Nature paper (covered by In Sequence 7/26/11) they detailed how they could shrink their features and increase the chip size to achieve first 165M and then 1100M sensors per chip. I've used this information, along with some really useful 'real world' feedback on the 314 and 316 chip from Nick Loman's blog (@pathogenomenick), to create a speculative roadmap for how Ion Torrent might achieve the 10x improvements through the middle of 2013. Assuming they are able to achieve the higher density chips in the timeframe I've listed (I don't think any timeframe was listed in the paper), that they can extend their reads out to 1000b (which seems reasonable given where 454 is today) and that they can improve the number of 'effective beads' (based on fill efficiency, # of 'live' beads, and # of reads passing the filters), it doesn't seem too unreasonable that they might maintain their '10x improvements every 6 months' goal out through the middle of 2013.

I also took a stab at what future chips might cost based on Rothberg's statement that they would hit a $1000 genome by 1/1/13 and a $500 genome by 7/1/13 (In Sequence 6/14/11). I'm assuming 90Gb per genome and $250 in reagents per run. (The prices listed are the initial prices at launch, but they've already demonstrated that the prices for 'old' chips may come down once the new chips are released.) Everything seems pretty reasonable until you get out to the (theoretical) 324 chip at $5000. Since Rothberg only predicted a 2x drop in price/genome from 1/1/13 to 7/1/13, they're either going to charge a lot more for this chip, or the 10x improvements will have started to peter out.

It should be fun to see how accurate this roadmap is. As always, comments and criticisms are welcome.

Chip1	Release Date	Output Spec	Sensors	Fill Eff. 2	Live Beads	Passing Filter	Read length	Calc Output	Est. Chip Price
314	01/01/11	10Mb	1.2M	41%	67%	67%	100	20Mb	$250
316	07/01/11	100Mb	6.1M	80%	67%	67%	100	200Mb	$250
318	01/01/12	1Gb	11M	80%	67%	67%	250	1Gb	$500
320	07/01/12	10Gb	60M3	80%	67%	67%	400	10Gb	$?
322	01/01/13	100Gb	165M	100%	100%	100%	600	100Gb	~$750
324	07/01/13	1Tb	1100M	100%	100%	100%	1000	1Tb	~$5,000

 

¹The chip names 320, 322, and 324 and release dates are purely speculative. I haven't seen Ion Torrent claim a new chip every 6 months, just that there would be 10x improvements every 6 months.
²The 'fill efficiency', 'live beads' and 'passing filter' metrics for the 314 and 316 chips came from Nick Loman. The values for the other chips are pure speculation on my part. The values for the 322 and 324 chips are surely a little optimistic, but they could be counterbalanced with longer reads to achieve the same output.
³The 60M sensor chip is also pure speculation on my part. Ion Torrent didn't mention this 'intermediate' chip in the Nature paper.