First a table scraped together from public sources:
Company: PACB , GNOM , ILMN
Market cap: $630M , $220M , $7B
P/E: n/a , n/a , 81
P/S: n/a , ~200 , ~8
Cash: ~$260M , ~$50M , $800M
Burn rate: $120M/year , $50M/year , profitable
Out of cash: ~YE2012 , ~YE2011, n/a
Proven technology: no , yes , yes
Commercialization risk: high , low , none
Commercial stage: not launched , launched , market leader
Current market share: none, low , very high
For those who have been in this business long enough, you know to put a premium on proven technology that works today. Far too much money has been lost on hoping that things "work out", especially when early slippage has occured and preliminary data look shoddy.
On this basis, I would personally only consider buying ILMN (though even it is quite expensive). As can be seen above, GNOM and PACB will almost certainly run out of cash within 1-2 years and require additional dilution.
I personally believe that PACB has a lot further to fall because they have totally unproven technology that continues to get pushed out time wise. GNOM and ILMN both have solutions that work *today*.
As I've written elsewhere, PACB appears to be in no man's land regarding specs (low accuracy, low throughput) with a very expensive box in an era where sequencing companies have already sold to the PACB intended audience. I just don't see who will fork over cold hard cash for this system when cheaper, higher performing alternatives are available.
Best wishes to all.
Interesting interpretation. And I agree with you that PACB is a risky investment as it is an unproven technology.
On the flip side, you could have said the same thing about ILMN back in 2003, which is when I started buying shares. There were huge price swings those first two years, but in the end the investment has paid off handsomely. I would not put all of my money into PACB, but I do believe in cost averaging some shares over time and see what happens.
Yes, extremely unproven -- even the company's numbers don't look very good.
Using ILMN is one comparable, but Solexa was a bold new technology in greenfield and took throughput to a level that no one before thought possible. Today, there are three big dogs in town (ILMN, Life, Roche) who have all sold over this marketplace. PACB is a step backwards in terms of throughput and accuracy, so I'll let the marketplace decide how they will fare.
My prediction would be that this will look much more than HLCS than ILMN. This could easily fall 10X in 2-3 years.
Only time will tell who is right...
You clearly have an agenda in your writing, altough it is very subtle:
PACB is sequencing a single strand. I'm not sure how large their "chip" is. And I say chip because i am not sure exactly what media they are sequencing on (ILMN-what I use) has a glass slide and Roche's 454 is more like a micro "plate" from my understanding. I am not sure how many single strands they are sequencing. I would throw out the number millions or more with confidence. From my understanding this is a highly scalable technology meaning that millions can go to ten millions and up very quickly. This means that if they have enough copies of the sequence the error rate can be lowered to as close to 0 as one needs to go.
I would say on average, we throw out 10% of our data that fails to pass our quality filters. This suggests that illuminas error rates are higher than what has been suggested by other posts.
In addition I don't think this has been thoroghly tested enough to determine the true per read accuracy. Since no machines have been tested in the field, this DNA is coming from either prospective purchasers or in house. DNA quality, source, and method of extraction play a crucial role in how well it sequences. Its possible that a simple change in the extraction method could completely change any of these numbers. I am doubtful that they have tried even all of the standard one (although I'm sure they have stuck to methods that would seemingly be the best for their platform).
Typically errors in reads are easy to detect and can be done fairly well with the right software. You see the same "artifacts" over and over again. You can detect them by where their location is in the read and if they are all in the same direction (I can explain later).
The only thing I don't understand about some of the posts I've read is why some people think its not high throughput. Not only is it already beating out the competition, later versions should be significantly better.
One more thing. Insertion/Deletions are very difficult to detect with 2nd generation sequencing technologies. I'm not sure what the PACBIO position is on their INDEL detection, but I would say that illumina fails horribly. As their kits get better their indel detection get better but it is still not that great. And with illumina the longer reads mean significanly more time. Its a roghly linear relationship. So you are looking at going from 2 days to 4 days in order to have detection of 20bp indels instead of 10. But with illumina you can forget being able to detect copy number variations at all.
emilio === thanks for taking the time to add some valuable insight about PACB -vs- ILMN.
As a very techie-BUT non-Life science person, it is extremely hard to read, interpret, and draw conclusions about the viability of our system -vs- any other sequencing system.
I look foward to hearing more quality comments about our outlook.
Emilio, you wrote: "I would throw out the number millions or more with confidence."
Misplaced confidence my friend. Check out what their Chief Scientific Officer says about throughput:
Check out p. R233.
"In addition, the throughput of SMRT sequencing
will not initially match what can be achieved by SGS [second generation sequencing]. The throughput of SMRT sequencing is a function of the number of ZMWs that can be read at once. While ultimately the potential exists to observe many ZMWs in parallel, the first version released will be capable of only up to 75,000 ZMWs."
So even the company admits that their throughput will be less than current technologies!
The question is how much worse will PACB be? The HiSeq is doing 300 GBs over ~ 10 days.
If we do some back of the envelope calculations, you get to pretty horrid numbers for PACB. Let's start with what the company says about occupancy (p. R321)
"At present, because the DNA polymerase and DNA template to be sequenced are delivered to ZMWs via a random diffusion process, approximately a third of the ZMWs of a given array are active for a given run."
Let's assume that they can do 500 bps at a run (more than what people can currently do, but let's be generous), and that they can do 100 runs/day. Then you get:
25,000 ZMWs * 300 bps * 100 runs = 0.7 GBs
Even if they can do 1000 runs / day, which would imply 1.4 minutes per run, you get to 7 GBs. This assumes that it is running 24 hours/day w/o interruption.
So at the very highest assumptions, they are still *well* below the 30 GB / day that ILMN delivers. This doesn't even factor in the much higher error rate that they are getting per read.
More realistically, we are talking about 0.5 GBs / day, so they are looking to be ~ 60X *worse* than ILMN.
So this clearly looks like a low throughput, high error machine.
Why would anyone buy this?
Deepanalyst, thanks for all the posts you really bring up solid points. Have you looked at gen-probe at all? if you have what do you think? I like this company, but I think strong legislation cracking down on the patenting of genetic sequences and arrays might cut into the genetic diagnostic business plan.
On a side note, I am weakly familiar with the Helicos system as it was used in my molecular anthropology classes during graduate school (I am by no means a molecular biologist). I really hope the PACB system does not turn into an expensive paper weight. The investment by Monsanto, makes me want to believe that PACB has real promising technology.