Miniature GC-MS: just one problem . . .

Another one of those stories about what is truly, a technological marvel: shrinking a gas chromatograph/mass spectrometer down to the size of an iPod, with the target size being that of a matchbox. Designed by MIT engineers, the device which can analyze the air for hazardous gases (and could be adapted for other media like water) is touted as a possible distributed sensor for water supplies to protect us against chemical attacks or in subway systems to warn of terrorist attacks. I think this is bullshit and I'll explain why after a description of this ingenious device:

Their detector uses gas chromatography and mass spectrometry (GC-MS) to identify gas molecules by their telltale electronic signatures. Current versions of portable GC-MS machines, which take about 15 minutes to produce results, are around 40,000 cubic centimeters, about the size of a full paper grocery bag, and use 10,000 joules of energy.

[snip]

The analyzer works by breaking gas molecules into ionized fragments, which can be detected by their specific charge (ratio of charge to molecular weight).

Gas molecules are broken apart either by stripping electrons off the molecules, or by bombarding them with electrons stripped from carbon nanotubes. The fragments are then sent through a long, narrow electric field. At the end of the field, the ions' charges are converted to voltage and measured by an electrometer, yielding the molecules' distinctive electronic signature. (Science Daily)

Sounds great, right? Especially if it is very sensitive, that is, it is highly accurate in detecting the hazardous material when its there, even in very small amounts. But therein lies the rub. Because to do this it is likely the device will suffer at least a little bit in specificity, that is, it will sound false alarms. It doesn't take much. Even if it is highly and unbelievably accurate in this regard, say, 99.99999% accurate, there is a big problem.

Let's do some back of the envelope calculations. Let's say a single small sensor is 99.99999% specific (i.e., its false alarm rate is only .0000001, one in ten million). This device is said to perform an analysis in about 4 seconds, which is to say it could, with ease, report a reading at least once a minute, something it would need to do if it were to continuously monitor the environment for a terrorist threat. Let's put it in a water distribution system. That's 3600 readings a day or about 1.3 million reads a year from that sensor. But a water distribution system is big and you would probably have at least 100 of these sensors in your water supply. So that's 130 million reads a year. Now most years there will be no terrorist attack on any water supply. There has never been such an attack and there are over 150,000 community water supplies in the US. But even at the rate of one attack a year (an absurdly high frequency) that means the risk for any water supply is one in 150,000 per year. If you insist that some supplies are more likely to be hit, say the top 1000 supplies, it is still a chance of only one in a thousand per year.

But how many false alarms will we have? For a single water supply (a supply that has only one chance in a thousand (at best) of being hit), the number of false alarms will be 13. So year after year this extraordinarily accurate sensor will spew out false alarms at the rate of more than one a month. How long to you think it will be before they are no longer maintained and in fact are just disregarded or disconnected?

These tech stories are great. It may well be that a tiny GC-MS will find an application somewhere but putting them into the subway system or our water supply isn't likely to be one of them.

More like this

We use small gas detectors all the time in Refineries, petrochemical plants, oil and gas production facilities etc. Yes false positives are a problem, but this is routinely handled by installing multiple detectors with consensus voting. The most common configuration is called "2003" or 2 out of 3 voting.

The Russians used poison in Afghanistan Revere. The wells are still being sampled there. OTOH if they didnt sample the water or air and something DID happen someone would jump up and say why didnt we do this earlier. How could it be possible that we are so inept and incomptetent in government. SAMP and NEST flights happen everyday over major cities in the US. Deterrent? No. But we sure as heck cant let our guard down as many would have us do.

Here is a low tech solution to your machine problem.
http://news.nationalgeographic.com/news/2006/09/06928-bluegill-fish.html

I can also tell you that it would be easy to take LA and Phoenix AZ out and they wouldnt know it until it was all over. There is one single material that was discussed in a briefing. As a result I only use my well water and/or distilled water for drinking. I wont say it here but its undetectable except by direct testing and only gives up a slightly higher reading than normal in lab testing. Chromatographs are able to pick it up though as are spectrometers. .

Sampling air and water supplies even if its giving up false alarms are better than no detection at all. Water is the most vulnerable system we have, followed by food. Blow it on the stuff that was touted for water and food and six months later it would make a pandemic of flu look like it had blown thru on the first wave. Long, slow lingering death and it would begin with a bad case of the stupids and blank stares at people. Later, you lose all muscle function. Diagnosis a lot of the time is Myasthenia Gravis but thats not it as they figure it out as they progress and do tissue samplings. You go with your eyes open and your CNS just shuts down.

Death is 100% assured.

You math is dead on. But in coordinated attack and just using the number of suspected people in the US that are mole terrorists right now they could in a week target over 10,000 systems, just by doing two per day. Air sensors in subways? Go for it. We saw what happened in Japan with the Sarin attack. If the little gizmo sends me running for the surface I will just be inconvenienced. If it sends me a lot I would say please get better ones but I will head for the surface every time. But if the assertion is that like a fire alarm in a building it goes off you would do nothing then maybe you should rethink that one. If one went off in a subway, then likely no alarm. You get five going off you hit the button. It would only save a bunch of people. Those that are in proximity to the first five, well they are toast anyway. But you did save the unknown number.

Yeah, I like false alarms. It means they are checking.

By M. Randolph Kruger (not verified) on 16 Jan 2008 #permalink

Imagine ten thousand of these in the NYC subway system. The teeming millions are teeming with odors -- body odors, soaps, deodorants, antiperspirants, lotions, unguents, cold treatments, perfumes, colognes, aftershaves -- and lots of them carry food with them, foods of a thousand cultures. Lots of bluecollar people ride the subway in their workclothes which have plenty of solvents in them.

When the gizmos breakdown chemicals and analyze fragments, the logic is making guesses as to what wasn't analyzed. This lack of specificity will result in countless false alarms. The first day or so, the response teams will not get panicky at every positive. Eventually, the unofficial policy will be to run the gizmos in the 'OFF' configuration, or else disconnect its telemetry.

Of course this result won't obtain until the vendor has sold 10,000 units -- and been paid in full.

Asking wouldn't-it-be-possible without asking would-it-be-practical is always a huge mistake.

By Globle Warren … (not verified) on 16 Jan 2008 #permalink

Of course this result won't obtain until the vendor has sold 10,000 units -- and been paid in full.

Or until they build a couple prototypes and try them out in actual conditions to see if they work...

road: the problem is that almost all reads will be negative and even tiny FP rates will produce so many FPs that the devices will be ignored or not maintained. The pos. predictive value is tied to the prevalence of true positives which in an oil refinery is orders of magnitude (maybe 5) more than in a subway or water supply. Clonsensus doesn't work because it sacrifices sensitivity for specificity and I've already set the specificity at almost 100% The difference between the real value and 100% is so tiny it is almost impossible to engineer but in the real world it dooms the application.

Randy: The math says that every time the alarm goes off it is a false alarm. Essentially every time. And we are talking about once a month. How long do you think it will take before it is ignored?

But you are making an assumption that it will do what you said Revere. Its a new technology given. But if it alerts us even once it will be worth it. That is unless they are horrendously expensive. I am also equally sure that there is a way to tune down the sensitivity or up as the case may be. Just like stealth bombers, there is a way to detect them and its not so sort of rocket science if you know what you are looking for.

I get the mental picture of the "seconds from disaster" segment from the Science Channel where the poor bastard being Japanese and so meticulously clean, bent down to wipe up the Sarin and had done a pretty good job of it, thus saving dozens if not thousands of lives that afternoon. I have seen what that shit does to dogs and mice. Its not very pretty and the re-enactment didnt cover it. If these things could false alarm as much as you say, then the go, no-go criteria would be enforced. We would lose a few but not a load. I dont understand your aversion to it other than its a gizmo that MIGHT inconvenience you. Do you disable your smoke alarms because cooking a steak sets it off? Hell people are less likely to panic if indeed it is a multiple false alarm system. The will move out slowly, quietly and once upstairs if its found the stuff is there then its served its purpose.

I personally dont mind taking off my shoes in light of the Richard Reed incident. There are thousands of ways to knock an airliner down, there are also as many ways to kill people on a subway as evidenced in Madrid, or a bus in London.

I guess my question would be something like at what level do you want them to stop invading your privacy or inconveniencing you and others. Have we reached a state of equilibrium now where we sit back and say, "We havent been hit yet so lets stop there?"

I dont know anything but what the security briefs put out , and ho-hum here comes another one kind of thing. Quite frankly I think we are heavily infiltrated and they are just waiting for the signal to go. If they are right...?

By M. Randolph Kruger (not verified) on 16 Jan 2008 #permalink

Randy: You miss my point. My point is that it is like a fire alarm that is constantly going off. After a while you disconnect it and you never put one like it in again. No water supply or transportation system will ever use one because they are too "expensive" in terms of utility. I know water systems and I've talked with the heads of some of the biggest in the world. They have said they would never use a system that gave continual false alarms and essentially never a true one (because the probability of a true one among the false ones is vanishingly small.

Kruger: If we are heavily infiltrated, the opportunities in America must be turning them to the light side pretty thoroughly. If there were a bunch of trigger happy crazy terrorists here running drills and keeping up their training, they'd make some mistakes, blow it and we'd have seen something. At worst they are low-lying moles working as cabbies and construction workers or whatever, not training, and probably doing a lot better by sending money home than they ever would by poisoning a city for the representatives of the sky-god. If they've sent terrorists slaves over here, they have found they've wasted their cash as the moles have gone native, and moved on to more effective things, like making scary videos.

The 'security' folk you are reading are puffing up the threat to inflate their worth.

Revere, I think *you're* missing the point about consensus. However high you set the specificity, you can get it higher by using consensus, and if you get it high enough, the device will be useful. You can't do it just by the numbers, it'll depend what circumstances produce false positives. For example, maybe the device turns out to be confused by brand X perfume. Then maybe your consensus should be between sensors which are close enough together that any attack would quickly set both off, but far enough apart that they don't both detect the same person's perfume normally. It's not obvious that you couldn't design a system that would work. (Cost/benefit is probably a bigger issue.)

By Mathematician (not verified) on 17 Jan 2008 #permalink

If the devices are sampling every four minutes or whatever, and a single false positive shows up slightly less often than once per month, I think getting two or more samples in a row showing as positive should be distinguishable from a single false positive.
That said, I think "terrorism" is the new "cancer", i.e. if you can't claim that your research might somehow maybe someday treat or cure cancer, at least you can now claim that it can somehow be used to fight terrorism instead and boost your chances of getting grant money. (Why, yes, I AM slightly cynical about this kind of thing, why do you ask?).
A handheld portable GCMS would be really spiffy well outside of the range of "detecting terrorist attacks", but of course those potential uses get downplayed or unmentioned because they don't bring in the grant money and attention like "terrorism" (or "cancer") does.

These devices are like early warning radar, they give the operator an indication that an attack is under way. They are not 100.00% reliable, but that's not really needed. Let's assume someone opens up a canister of VX in the subway. You have X dead people, having the sensor does them no good. But, the system then indicates that VX was the agent and workers can then respond with the correct protective gear and hopefully antidote (assuming a nerve agent).

So Revere, you would rather just bag it and figure out what happened later?

I don't see any insurmountable problem. GC/MS has very fine resolution and is good for picking molecules out of a mixture - for example, it can be used for picking out specific drug metabolites in urine. It would be difficult to catch molecules that are structurally similar to benign ones, but it would be very useful in cases where the distinguishing features would shows up distinctly on GC/MS. For example, a compound coming out of the GC phase that shows up with MS peaks corresponding to an organic backbone and lots of nitro groups will almost certainly be an explosive.

It's possible to have a false positive problem would occur if they were deployed poorly (overreaction to positives with no confirmation, too low of cut offs for positives, placed in conditions where they couldn't operate reliably, asked to make distinctions that the technology is poorly suited for), but if they were deployed at appropriate points with appropriate thresholds, such as at high-risk targets or bottlenecks for people and materials going in or out of a contained system, they'd be effective.

For water supplies, it would probably work just fine - there are plenty of bottlenecks in the water treatment process to sample from and the end results should be relatively pure chemically so it should be easy to pick out if it has been contaminated. I doubt they'll be deployed at regular intervals in transit systems, because the cost of setting up enough devices to do it wouldn't be worth it, but it would likely find a role at security checkpoints where analytical insturments that are less specific than a full GC/MS are already being used.

Many decades ago I used to work on a lot with GCMS and it's been fascinating to watch them go from huge manual devices to automated microwave-sized ones. However, I have to agree with Revere here. It's okay to program a unit to detect a specific gas and its level. For terrorism purposes there are so many potential bad actors it's not a good method. Moreover, has anyone done a cost-effectiveness study? I'll bet not.

I had occasion, at one point in my life, to interview a number of Gulf War I vets about, among other things, their experiences with the chemical attack sniffers in use in the arena at that time, which were sensitive but not specific. Their reports to me were consistent with Revere's observation about faulty fire alarms: they stopped taking the alerts seriously, although since the soldiers had orders, they still donned their protective equipment.

What was sort of surprising, though, was that even though the vets reported not taking the alarms seriously after the first few days, they consistently reported that the alarms were a source of considerable stress. I inferred from some of their statements, too, that the repeated alarms followed by all-clears decreased their confidence in the systems and in the honesty of the higher-ups. That's the worst possible outcome: an alerting system that dulls people to the sense of urgency while making their lives miserable, and that causes the people it's supposed to protect to question the competence of the protectors.

If you want to protect the subways, there are known techniques that would do a lot, HEPA filter the air, irradiate it with UV, pass it through activated carbon.

Does that stop everything? No, but it stops infinitely more than a zillion sensors hooked to alarms do. It would also reduce the spread of things like flu.

Hang on now. Revere makes a couple of valid points. Deadie HEPA would stop bird bugs and that sort of thing, but not nerve gas. NGA's actually eat up chem suits pretty quickly and it contaminates the shit out of them. VX for example if it contacts the skin in any way you might as well be ready to send for the wagon. Basically to get rid of the agents you have to burn them and it renders the agent fairly inert....

I personaly believe that if these things dont cost 10 g's apiece that it would be worth it. I am assuming that they can set them for explosives, chemical agents but obviously not bio bugs. If they are used then they would likely be deployed as consensus sensors rather than single point sensors. I am also equally aware that the might be prone to decalibration. But these are things that are easily fixed and cheap to do.

One of you emailed me about what the substance was that would cause 100% death. Give it a week or two and it will probably come out in the news anyway. This stuff is wicked and I cant tell you until it breaks in the news. Probably be out this afternoon the way things come from Leaky Leahy's office.

By M. Randolph Kruger (not verified) on 17 Jan 2008 #permalink

These tech stories are great. It may well be that a tiny GC-MS will find an application somewhere but putting them into the subway system or our water supply isn't likely to be one of them.

This is a crucial component to a functional tricorder.

stu et al.: You are all missing my point. It isn't that I think it wouldn't be good to know if there was an attack. It is that every time the alarm goes off it will be a false alarm. Essentially the mathematics are that every alarm is a false alarm. Such a device is useless and will not be used. If an alarm gives a false alarm once a month for an average of a thousand years, it won't be used.

Mathematician: Consensus helps specificity (but how much help to you want for 99.9999% specific?) but it spoils the sensitivity. The problem, once again, is the expected rate of a true signal is so extremely tiny that almost every reading is one of no true signal so it doesn't take hardly any false positive rate to overwhelm the true.

"This is a crucial component to a functional tricorder."

ROTFL!

By machinist (not verified) on 17 Jan 2008 #permalink

No, I think you are missing the point of the program. It is not to be 100% accurate, or stop an agent release. It is to report what is there. Like I said, once the agent is released no detector can do anything about it. It is however very important to know what was released so you can respond accordingly. Just like EW radar, you don't launch a strike based on an indication, it gives you a warning that there may be a problem and you need to make additional checks.

I do happen to know a lot about these systems, I can't say how but there are versions that don't use the traditional GC technology that have essentially a consumer electronics cost structure. So don't assume the entire R&D effort is "bullshit", there is a lot you don't know.

MRK: Can you at least say a) rough quantity needed to render lethal a quantity of water going into a distribution system for a city of 1 million: a test-tube full, a suitcase-full, a truckload-full, or more, or less? And can you say if there are realistic measures for a) detecting it, and/or for b) neutralizing it, and/or for c) treating people who have had contact with it within a specific time window?

---

Revere, I have to disagree with you on this one. Empirical testing can lead to the design of configurations of multiple sensors that minimize FPs. Water supplies and other limited systems can benefit. And I for one would rather put up with false alarms from devices of this type, than put up with ubiquitous surveillance of the type that has made the UK an Orwellian parody. Inconvenience doesn't kill.

---

Oh my, newz on the radio at this moment: Florida, 83 cases of Norovirus....

The unquoted bullshit part of the referenced post that Revere takes issue with is "The device, which the researchers plan to have completed within two years, could be used to help protect water supplies or for medical diagnostics, as well as to detect hazardous gases in the air."

I think it is pure sales and funding stuff and agree with Revere about the reasons why. In real life, we'll end up using these things as diagnostic tools after the fact, and they'll speed the response, rather than alert tools before the fact. Our emergency managers have far too many false alarms to conted with as it is, and adding possible chemical attacks will fall below possible heart attacks in our e911 priority tree.

This technology may enable a mini-GC-MS in every ambulance, and actually help diagnose what went wrong, but the real detectors will still be people. The probability problem will be more like: given that these people were affected, was there something in the air, water, or food that did it?

Tricorder-wise, if we set our tricorders to alarm on noticing any danger, they'd never shut up and we'd never get anything done, so we'll always need to let a few red-shirts die. The better tricorder technology will help us figure out what happened.

stu, g234: It turns out I know a reasonable amount about this topic (I am cleared to know and do). Please read the calculations I made again. I am assuming 100% sensitivity and 99.9999% specificity. You can't do much better than that. What is wrong with my calculations? That's the issue. Please critique the analysis, not just disagree.

DaveX: Yes, that's the part I am objecting to. Should have included it.

These tech stories are great. It may well be that a tiny GC-MS will find an application somewhere but putting them into the subway system or our water supply isn't likely to be one of them.

Sounds like a nice device. Even if it can't work on the subway or water right now, maybe further research will lead to something that can?
Dave Briggs :~)

MRK, activated carbon will take out nerve agents.

The problem with any sensor based approach is as Revere points out the false alarm problem, and even if you know exactly what it is, so what? Then what do you do? Beam all of the victims into sick bay where medical technicians are ready to inject the right antidote (if there is one) into the victims?

Suppose you want to "detect" a cloud of toxin that is 25x25 meters. How many sensors do you need in that area to get reliability? How many 25x25 meter areas do you want to protect? How many sensors is that? What is the installed cost of the sensors and the infrastructure to keep them going? That includes the communication links, the power sources, the technicians to check, calibrate, service and repair the sensors? Don't forget the communication links to the responders, and the prior training of those responders so they know what to do with the information they might receive.

How many nerve gas attacks have there been on civilians in the developed world? I am only aware of one, in Japan. If there had been such a system in place in Japan how many lives would it have saved during that attack? In that attack, 12 people died and 50 were seriously injured.

http://en.wikipedia.org/wiki/Sarin_gas_attack_on_the_Tokyo_subway

How many lives would such a system have to save to justify spending money on it instead of spending money on something like universal health care? Remember the lives saved in a hypothetical gas attack are hypothetical, the lives saved in incresed health care are not.

The main thing this article tells me is that it identifies an engineering challenge. The first step in solving a problem is to be aware of its existence. Obviously a device that can detect hazards materials in the water supply would be of considerable value. Instinctively a GCMS sounds like a good solution but the analysis in the article illustrates that current technology is not up to the task. I view this as a necessity in search of an invention. Television cameras initially cost hundreds of thousands of dollars and were huge, now they are tiny and cost a few cents. The same thing is likely to occur with trace particle detection systems if the need is there. the GCMS approach is just the first step in finding a series of engineering solutions.

By ProfChuck (not verified) on 17 Feb 2012 #permalink