- One Cobas 8800 is able to perform 4100 tests per day in a fully automatic fashion.
- There are roughly 800 Cobas 6800/8800 in Europe, US and Asia combined.
- One test costs about 15 CHF (which is roughly 15 USD) (EDIT: This is what Roche gets per test. The overall laboratory cost is probably much more expensive).
- Roche is pretty much the only manufacturer for machines like these.
> Roche is pretty much the only manufacturer for machines like these.
Though Roche makes the only huge machines like this, there's lots of high-automation rtPCR out there. And you can fit a lot of e.g. Abbott m2000's in the floor space that one Cobas 6800 takes, picking up some flexibility and each doing ~500 a day.
That is a very well produced video, it doesn't contain a single spoken word and yet it tells the story perfectly. Incredible machinery. Any idea what these cost?
Yes, and if you install one on site, they send dedicated techs to set it up, which lasts a month or more, plus custom integration with site software. I worked with a similar machine and if it lost its net connection to the backend database, it would dump half a million $ in reagents down the drain when it entered "safe mode".
The video above (at least one of them) was showing an entire automation line built on Roche technology-- easily an order of magnitude more than 400k. The Cobas 8800 is just a component of that entire line.
Ah ok, that makes more sense. I figured a 100K or so per unit would be cheap the way it has been engineered. That's a whole robotic laboratory and what with the normal price of cleanroom equipment it would be ridiculously cheap if that setup was a few hundred K.
Impressive video and soundtrack, indeed. One thing that I don't fully understand, though, is how relevant lab results might be accurate, when, during part of the processing workflow, containers appear to be open to the processing chamber and its atmosphere, thereby presumably being exposed to potential cross-contamination. I'm likely missing something trivial, so I'm curious about what it might be.
I guess it might be flushed by cleaned air or nitrogen ?
But good point about cross contamination, still could be there is a mechanism to counter that, like some vapour pressure tricks or gas streams preventing fumes from one open container reaching another.
I worked on the ancestor to these machines over at Abbott. There’s nothing in whole blood or serum that will spontaneously aerosolize and get into the air.
The tests are detecting levels of antigens, antibodies, or DNA in the sample. The samples are manipulated with disposable parts to prevent crossover.
That is not huge - in fact that looks like there's an incredible amount of miniaturization already put into that product.
Technology is great. Before this product it would probably take an entire university campus with multiple labs full of technicians to do the number of tests this is doing.
Researching the automated MDx market, I found out the Cobas 8800 (and indeed, pretty much all automated MDx solutions) only works with Roche-manufactured proprietary panels, or results are only valid with proprietary panels. BD lets you run user-originated panels, but you can mix them with FDA-approved panels. I can't just take some master mix from someone else and generate a valid result on a Cobas 8800, as far as I can tell. Would appreciate any clarification/correction on this, though.
What is Roche's manufacturing capacity for these panels each week for the US in the next four weeks? With the 110 machine figure someone else reported in this thread, there is a theoretical limit of about 451K Roche tests per day assuming we shove aside all other testing on this gear, but we could be supply-constrained by these proprietary panels instead. In which case we really should get BD MAX, Seegene All-In-One Platform and similar S. Korean solutions' approvals from the FDA specifically for their Covid-19 panels for their systems on an national emergency basis.
Considering this is HN, there is enough engineering brainpower to build open source clones of this. Biology is not the same as electronics, it is closer to software in the sense that a lot of things can be easily (for some definition of easy) cloned once you figure out how it works. The only difference is that biology is slow and expensive and takes a lot of equipment.
Caveats: getting to Roche level of throughput and scale would be tricky with just Arduinos and off the shelf stuff and there are some magic numbers that you will have to figure out through experimentation. Building a usable machine just for a small number of people is not that tricky if you have e.g. 5 bio and 5 CS/EE people working on getting a design shipped.
I think you greatly underestimate how complicated these machines are. For instance, I just cannot imagine how these machines keep themselves so clean. My guess is "lots of disposables" but still, that must take years to get right. Just read "Bad Blood" to see how Theranos struggled with the machines not simply becoming a bloody mess, and they had some great engineers on the team.
Look at the FDA testing protocol [1] for their Covid-19 test kit.
They use molecular water (double-distilled, deionized, likely de-salted water), and DNA and RNA degradation solutions, some PPE, autoclaves, bleach, and lots of workflow procedures to carefully mitigate cross-contamination.
I'm curious why the life sciences folks aren't standardized upon something like semiconductor SMIF to isolate the samples and point of contact of equipment into a modular sterile environment, and limit cleaning to interface ports optimized for sterile operations and maintenance. My first swag at such interface ports might look into some design that can be clamped and autoclaved, and submerged in some degradation solution, for example. Generally, the current open standard is terrifically manual labor intensive and prone to mistakes.
All the automated solutions I've been able to read up on so far (BD MAX, Seegene All-In-One Platform, Roche Cobas 6600/8800, BioFire FilmArray Torch and FilmArray 2.0) seem to lock in customers to proprietary panels. I'm curious whether that is more of a commercial decision or a scientific decision. Of these, only BD MAX allows user-designed protocols at all.
So far, I've yet to see a persuasive argument that an automated molecular diagnostics system must use a proprietary platform for scientific reasons. A generic automated platform that can use anyone's glassware, molecular water, primers, master mixes, etc., will need many more sensors for quality monitoring, and those need constant calibration, so there is a significant engineering tradeoff. But I'd be surprised if any of the proprietary solutions are much more cost-effective; it seems all of these kinds of solutions are following the John Deere lock-in servicing model to generate profits.
The lion's share of the test volume is done by commercial labs like LabCorp and Quest, who don't care that much about designing their own tests, they just want to maximize patient volume.
One of the companies I used to work for had a large chemistry system that could be run "open" but only a few customers used it that way, a couple research laboratories and I think a winery. Thousands of other instruments were used for plain ol' medical testing.
Even if you designed a sufficiently open system that could run Roche or Siemens reagents or whatever, any problem with the reagents and the company will just shrug and say "it's not our instrument, no guarantees."
It's mostly a commercial decision. Reagents have very high margins and recurring revenue is more attractive than large one-time purchases. Illumina does this with genetic sequencing as well.
edit: There are some informal standards like 96- and 384- well plates that make it possible to have third-party consumables with some machines, but there is no incentive for a for-profit company to design open platform.
These folks [1] are working on an open Covid-19 test. There is a pretty significant logistical train that I haven't seen systematically addressed yet in the project, though lots of those pieces have open projects that can be picked up, adapted and integrated. Even failing will be valuable: that identifies gaps to address.
I've done IT for labs. The PCs these insanely expensive machines are hoked to make me absolutely sick with how they have to be set up. You have a testing machine, be it a mass spec, PCR, whatever, costing 6 and 7 figures with a $500 PC that has every imaginable security hole and updates turned off. Drives me bonkers.
That PC is only for controlling this machine and no need to be on internet, right? If so, just config the windows firewall to disconnect it from the internet?
I do that on my home machines. All updates are disable and I manually enable them every 6-12 months to let win10 to go crazy for a day or so and everything is up to day.
One can also monitor the podcast such as "Security Now" to check if the latest windows update processes have any special BSOD or silently / randomly delete all users files/profiles before turn on the update.
Except you need them on the local LAN to share data, and you need them to have internet access through the local LAN so the vendor can log in remotely for maintenance and troubleshooting.
Designed to provide a fully integrated real-time PCR workflow,
the cobas® 8800 system can run up to 960 tests in an eight-hour
shift with minimal user interactions and up to four hours of
walk-away time per run when running at maximum capacity.
It looks like that the 960 number is predicated upon a start time from "idle", and minimal interactions. The actual slope rate of test completion looks like it's 192/hour. If that throughput can be maintained with more frequent interaction to keep the machine fed, it'd be 4.6k/day.
https://www.dialog.roche.com/content/dam/dialog/owp/apac/Med...
That is, it takes time to "stuff the pipeline".
The cited number -- 4128 -- appears to be 21.5 * 192.
This South Korean manufacturer seems to able to do 94 tests at a time in four hours with their machine. Which might be a nicer batch number for testing closer to the point of care.
I think this is similar to what everybody else is doing -- using a thermocycler to run PCR. It can run 96 samples in parallel and takes 3-4 hours to run through 30-40 cycles. But there are a lot of manual steps, and I'm not sure how much the SK company automates them.
This can't be the only option for testing. Korea has tested hundreds of thousands of people so far. By far the largest number of any country, including China. Such high throughput does not seem achievable with these machines unless you already have a ton of them, and all the reagents are stockpiled.
Everyone else seems fairly low throughput, with the US being on the "ridiculously low" end of that spectrum if you consider tests per million of population.
SK does 10.000 tests per day with test kits which they could quickly produce but which involve a lot of manual steps and hence man power. You could replace all that with 2.5 of these huge Roche machines. So for really high throughput you absolutely need those huge machines, everything else simply does not scale.
Of course you need a stockpile of reagents, but manual test kits need these as well and AFAIK those are fairly standard.
There are 110 in the US (of either type; most are likely the 6800), and Roche says they've installed a "significant" number in the US in the last few weeks (who knows what that means).
These are hardly the only machines that can run rtPCR though; they're just highly automated.
eg biorad makes and sells tons of (much simpler), typically 96-well rtPCR machines. You can find them on ebay for like $5k. Training people to use them is not particularly hard, and you can probably do 4-8 runs / day on one of them. NB: 96 will means you probably can process 88-92 samples per run plus 4-8 controls.
The limiting factor is approved tests and reagents, laboratory bench time to prep samples, and a distribution mechanism to get test kits to to people and then to labs.
- Roche has been working on this already for weeks with high priority.
- These are tests that run on high-throughput machines, namely the Roche Cobas 6800/8800 systems:
https://www.dialog.roche.com/pk/en_us/products_and_solutions...
- One Cobas 8800 is able to perform 4100 tests per day in a fully automatic fashion.
- There are roughly 800 Cobas 6800/8800 in Europe, US and Asia combined.
- One test costs about 15 CHF (which is roughly 15 USD) (EDIT: This is what Roche gets per test. The overall laboratory cost is probably much more expensive).
- Roche is pretty much the only manufacturer for machines like these.
Source (German):
https://www.faz.net/2.1690/pharmakonzern-roche-neuer-schnell...
EDIT: More exact numbers in the linked article: Cobas 6800 can do 1440 tests/day, Cobas 8800 4128 tests/day.