Slashdot Mirror

← Back to Stories (view on slashdot.org)

Startup Uses Sensor Networks To Debug Science Experiments (xconomy.com)

Posted by timothy on from the watching-the-watchmen dept.

gthuang88 writes: Environmental factors like temperature, humidity, or lighting often derail life science experiments. Now Elemental Machines, a startup from the founders of Misfit Wearables, is trying to help scientists debug experiments using distributed sensors and machine-learning software to detect anomalies. The product is in beta testing with academic labs and biotech companies. The goal is to help speed up things like biology research and drug development. Wiring up experiments is part of a broader effort to create "smart labs" that automate some of the scientific process.

25 comments

  1. Disrupt the planet! by Anonymous Coward · · Score: 0

    So a bunch of hipster college dropouts in San Francisco are going to save the scientific method from SCIENTISTS with IoT and DSP/Machine Learning hocus-pocus?

    This should be entertaining!

    https://en.wikipedia.org/wiki/SCIgen

  2. Who needs a startup? by Sir+Holo · · Score: 1, Insightful

    I'm in the physical sciences, and even there am met with continuing reluctance of graduate students to take thorough lab notes in a lab book.

    It is not that hard to write, "It's humid today," or whatever. No matter how mundane the variable is, and no matter how fucking smart you think you are –with your imagined ability of total recall even a few months after the lab-time, everything is worth writing down.

    That way, when an anomalous result appears, they can search their notes for possible causes. Instead, they spend their time on FaceBook while the expensive instruments spit out Results – Results which all-too-often have inexplicable scatter in measurements of the variable-of-interest.

    BTW, I teach at a Global top-10 Sci-Eng University. The grad students' 'arrogance issues' seem to increase the further up the chain of Universities one goes. These kids resist direction like mad, and as a result, will never become world-class engineers or scientists.

    1. Re:Who needs a startup? by Anonymous Coward · · Score: 0

      Just give the stupid kids impossible assignments and fail them all.

    2. Re: Who needs a startup? by Anonymous Coward · · Score: 1

      Think about your own words: could this be the result of somebody trying to make us all more and more stupid? If that happens on a world class Univ, think about down the list ones.
      I dont think this is an accident.

    3. Re:Who needs a startup? by Anonymous Coward · · Score: 0

      ''I'm in the physical sciences, and even there am met with continuing reluctance of graduate students to take thorough lab notes in a lab book.It is not that hard to write, "It's humid today," or whatever. No matter how mundane the variable is, and no matter how fucking smart you think you are –with your imagined ability of total recall even a few months after the lab-time, everything is worth writing down.

      Experiment: Oil Drop Experiment
      Date: Monday 8 January 2016
      Time: 09:35 GMT
      Temperature (K): 294
      Pressure(torr): 764
      Gravity (m/s): 9.7
      Average Magnetic Field strength (G): 0.49
      OFC cables used: yes
      Coffee: present Type: Arabica
      Black Cat spotted: yes Lucky: yes
      Red Haired colleague in lab: yes
      Moon in the Seventh House: yes
      Jupiter aligned with Mars: no
      LHC Operational: yes Distance of experiment from LHC (m): 1000000
      Number of R's in the Month: 2
      Chinese food eaten in the past 24 Hrs: yes Chinese: no
      etc, etc, etc.

      Obviously, any results this one gets should be taken with a large pinch of salt..

      (Btw,

      ' continuing reluctance of graduate students'...'grad students' 'arrogance issues' seem to increase the further up the chain of Universities one goes'..

      been there, done that, worn out several t-shirts...)

    4. Re:Who needs a startup? by Anonymous Coward · · Score: 0

      Aye - tis a troubled lot this millennial scrum - grew a couple myself - not sure how that experiment will turn out. Incomplete data.
      Whether in simulation or just processed measurements - the question that I ask - and try to get them to ask - is "how do you *know* what you know?"
      Hint - the answer is *not* that is what the instrument showed on the display/simulation/graph

    5. Re:Who needs a startup? by Anonymous Coward · · Score: 0

      "The children now love luxury; they have bad manners, contempt for authority; they show disrespect for elders and love chatter in place of exercise. Children are now tyrants, not the servants of their households. They no longer rise when elders enter the room. They contradict their parents, chatter before company, gobble up dainties at the table, cross their legs, and tyrannize their teachers."
      https://en.wikiquote.org/wiki/Youth#Misattributed

      Young people have always resisted instruction. Particularly from teachers that have failed to secure their respect. Your best bet is to make up some war stories and funny anecdotes about saving Elon Musk from a model rocket accident or something.

    6. Re:Who needs a startup? by ganv · · Score: 2

      Yes, they do need to record good notes. But the times are changing. There is huge potential in a cheap general purpose sensor array. Imagine a system that recorded temperature, humidity, room brightness, vibrations, maybe some chemical species concentrations, and maybe data from a set of special sensors in an apparatus. This is straight forward to do, but right now it requires someone to write custom software and integrate a variety of sensors. If there becomes a standard that just ran with very little customization, it could greatly decrease the amount of 'something happened to this experiment but I don't know what'. It is easy to say 'they should take better lab notes', but no one has ever been able to accurately record all the relevant conditions. It is an imaginary world we describe in which the conditions of the experiment are recorded in the lab notebook. In research, you always have factors that you don't yet know are important, and high quality cheap general purpose sensors could be a big benefit.

    7. Re:Who needs a startup? by John+Allsup · · Score: 2

      As Richard Horton put in his essay (What is medicine's 5-sigma), 'As one participant put it, “poor methods get results”'. There is pressure for 'measurable progress'. This has a number of nasty side effects. Things which do not lead directly to publishable results fall by the wayside, and things which serve no other purpose than to potentially explain desirable results as experimental errors, again, offer little which would result in 'measurable progress'. More and more, career scientists are being forced to chase jobs and funding, and to either produce what will yield future funding, or find a new career. Somebody who takes three times as long to produce less remarkable results, and at greater financial cost, will most likely struggle for work. Somebody who does enough to get published, and gets twice as much published in the same time, and costing less, will appear to be better to financial pen-pushers. This places a negative pressure on standards.

      --
      John_Chalisque
    8. Re:Who needs a startup? by mikael · · Score: 2

      "It's humid today" isn't exactly a scientific measurement. I'd expect some kind of electronic measurement with barometer, thermometer and hygrometer. You can get an all in one wireless system with automatic logging for less than £40. If you want to splurge out on a bit more (£130), you can get a wireless weather station that connects via the internet to a smartphone. There are probably others with more features and functionality, but it was the first I found.

      --
      Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
    9. Re:Who needs a startup? by Anonymous Coward · · Score: 0

      http://www.cnet.com/products/n...

    10. Re:Who needs a startup? by Anonymous Coward · · Score: 0

      I remember such reluctance when I was in grad school, but for our group it wasn't due to lack of awareness of how important such extra data could be. The complaints were not about the importance of the data, but about the importance of such data in handwritten lab books. The professors made it clear that many came from a different environment, where lab note books were stamped, signed off on, and checked in every night, but that was for IP purposes. In a time when a large chunk of the direct data is already recorded digitally for hundreds of runs a days, attaching electronic notes and indirect data seems to make a lot more sense.

      For example, consider looking at the effects of humidity. If people wrote it down in their notebook, and you have more than one person working on the experiment, to check for correlations would require getting multiple people to go back through all of their notes and do a lot of imprecise data entry. The time the researchers spend doing that is more than the cost of a humidity sensor. I've been on an experiment where someone was able to put together a script to scrape weather info from online for the past year in less time to check for one issue that we thought might be weather related.

      Once I got past grad school, I came across experiments that seemed to do things right in a much more modern way. There was still a note field for every run that could contain whatever extra info you wanted, and you can later search through all runs for keywords. All sorts of extra data was automatically recorded, so that the researcher count spend their time doing things a computer could not. There were places things forced the user to check that the sensors all had sensible values before proceeding with runs. Other newer setups automatically did a long list of checks to make sure sensors were working ok, and could email the person in charge of a sensor when they were not. The user could set flags to indicate when they were doing something weird, so that some data may not make normal sense.

      For setups like that, I don't think we need more sensor packages, as there are already a lot of automated sensors available and others that can be adapted by an undergrad hourly in short time. The hard part is the data repository, as most projects seem to use a heavily custom setup, even if there is some standard library buried below (many of which suck...). Places that made it too hard to add new data columns basically resulted in more likely chance something didn't get recorded, or got recorded in a generic notes field that was hard to parse later. Other places made it take only a minute or two to drop in a new variable, which could be anything from a manual entered value, a new sensor, a value scrapped from a website like weather, to things like webcam images.

      Anyway, there are problems with arrogance all around, both from the young and the old, and both are constantly insisting it is the other side that is the problem. Learning how to deal with that is a big part of what makes a person a good instructor. Just telling someone what is the right way is not teaching, but showing them so they see why (and making sure there is a good why in a particular case) is.

    11. Re:Who needs a startup? by Anonymous Coward · · Score: 0

      The components are becoming more available now, and as more tech-friendly people wind up in "wet sciences" labs the need is getting more widely recognized. My group put a basic system together (and everyone is invited to contribute additional sensor interfaces etc!). It's built around a Raspberry Pi and sensors from Phidgets (and note the Phidgets people sell a 4-20 interface so it can talk to all sorts of things).

      In terms of capability this is not new, but we're trying to automate the setup as much as possible, so that people who don't consider themselves "computer people" can set something up, run a couple of basic sanity checks on the outputs, and have a functional monitoring system.

    12. Re:Who needs a startup? by RockDoctor · · Score: 1
      ... and he misses the point.

      Plainly, from the scenario, measuring the humidity wasn't part of the original experimental plan. The experiment is already running, and what the lecturer is saying is that (some of) his students don't conceive that there might be something worth recording that isn't in the experiment plan. Realising that your plans may be wrong is the first step. THEN you go on to "well, what can I do about this.

      You'd also be able to (probably) tell if there were a humidity effect by doing parameter-free ANOVA on your existing data, or attempting to back-estimate the humidity on other days of the experiment, in order to determine if there is an effect, if it's large enough to be detectable, and if it's large enough to be worth the £40 tool, the £130 tool, or simply taking a humidity report from the weather website.

      --
      Birds are not dinosaur descendants;birds are dinosaurs, for all useful meanings of "birds", "are" and "dinosaurs"
    13. Re:Who needs a startup? by RockDoctor · · Score: 1

      You've never had to install sensor cabling, have you?

      --
      Birds are not dinosaur descendants;birds are dinosaurs, for all useful meanings of "birds", "are" and "dinosaurs"
    14. Re:Who needs a startup? by RockDoctor · · Score: 1
      [In voice of some millennials]

      That way, when an anomalous result appears,

      What is this thing you call an "anomalous result"? This can't happen to me. That's implying that I can't see an obvious problem before it happens, and that CANNOT be true.

      [end outraged voice]

      One of the things that you learn with experience is that you can actually be wrong. It's one of the things that a lot of people these days have to actually learn, because they haven't learned it in their pre-teen or teenage years.

      For the last several years I've been introducing "Bright Young Things", recently recruited to a major company to work in managing the acquisition of data from the Real World. They too, despite being bright people, have to learn that they don't know everything, and that the Real World has things going on that they don't know about, and don't understand.

      It's an education for them.

      --
      Birds are not dinosaur descendants;birds are dinosaurs, for all useful meanings of "birds", "are" and "dinosaurs"
    15. Re:Who needs a startup? by Sir+Holo · · Score: 1

      "It's humid today" isn't exactly a scientific measurement. I'd expect some kind of electronic measurement with barometer, thermometer and hygrometer. You can get an all in one wireless system with automatic logging for less than £40. If you want to splurge out on a bit more (£130), you can get a wireless weather station that connects via the internet to a smartphone. There are probably others with more features and functionality, but it was the first I found.

      You prove my point. Thanks!

      It's facile to record either qualitative info, or to auto-record quantitative data from your humidity-loggers. Either way, when confusing experimental results appear, neither of these resources is likely to be used.

      My point was to gripe about the lazy technique among many new grad students of late. It is a product of:
      * Our undergraduate programs automating chemistry-lab experiments to be 'push-button' easy.
      * Calculus homework consisting of typing a problem into a computer program, and printing the answer it gives.
      * Statistical Process Control (SPC) apparently is not taught to undergrads

      I don't have the time to teach them everything they should have learned as undergraduates. Grad school is supposed to be the next phase... Learning at any stage is hands-on and primary-sourced. Discussing how US Universities got here would be a long conversation. 20 years ago, Americans had a reputation overseas as being high-level technicians, and rightly so. The problem has gotten worse.

    16. Re: Who needs a startup? by Sir+Holo · · Score: 1

      Think about your own words: could this be the result of somebody trying to make us all more and more stupid? If that happens on a world class Univ, think about down the list ones.
      I dont think this is an accident.

      I said nothing of the sort.

      As a matter of fact, I've consistently found the focus on actually teaching at (your) "down the list" schools to be much stronger for undergraduates. Smaller classes. Professors actually teach their classes. And they grade students' homework. They will run the labs. With less-expensive equipment, students will have to learn lab-work by "Doing it the hard way," which is more effective.

      At a major Research University: More than 50% of classes are not taught by Professors, but by "Lecturers" – PhDs who take on that load so the tenure-track Profs. can bring in more grants – or post-docs who have not yet learned to speak clearly. Core classes can reach class-sizes well over 500 per section. An army of TAs ensures that students can't reach the Professor, who might schedule 'Office Hours' at a very inconvenient time. (I've done it – on advice.) (OK, but when I'm Advising grad students, I force them to "Do it the hard way," despite having software on my desk that could spit out an 'answer' in seconds.)

      The above is a broad-brush generalization, but holds true often enough. It mostly depends on whether a College or University cares about more than just money and football. Or both. Variation among Departments and Schools is a factor, too.

      So, students must research and also visit potential schools before committing. Caveat emptor!

    17. Re:Who needs a startup? by Sir+Holo · · Score: 1

      CORRECTION

      20 years ago, American PhDs had a reputation overseas as being high-level technicians, and rightly so. The problem has gotten worse.

  3. IF you Need this, then Experimental Design FAIL by Anonymous Coward · · Score: 0

    If the results of your experiment were significantly changed by the surrounding air's humidity, the ambient temperature, or a breeze in the room, then the experiment was not designed at all correctly. The whole idea of designing an experiment is to focus the perturbations on a small set of variables AND eliminate the other confounding variables. If you can't design an experiment, no gadget or start-up company can save your ass.

    1. Re:IF you Need this, then Experimental Design FAIL by KermodeBear · · Score: 2

      While true, this kind of effort can help scientists discover things that they have overlooked.

      I know that it is difficult to believe, but scientists are generally human and they're not omniscient. Their experiments often have some kind of problem.

      --
      Love sees no species.
  4. About time... by virtualXTC · · Score: 3, Interesting

    This sort of thing has always been available for pharmaceutical manufacturing, but has been long overlooked on the research side. I've been at a few science research based engineering companies that collect this kind of data already, but don't do anything to analyse it unless something catastrophic happens. A software tool that could enable visualization of this data across experiments will extremely valuable as we remove technician to technician variation (via robotics) and a synthetic biology becomes more common place, and could prove as invaluable as well plate edge effect analysis already included in major bio-analysis software packages such as spotfire.

  5. Benefits seem pretty contrived by Dr.+Spork · · Score: 1

    I read TFA and it struck me that this is the invention of salesmen who are working very hard to find a rationale for their product. The two examples they came up with, where the benefits of their system are supposed to be maximally evident, are just not convincing. In the case of the mice who are kept awake at night: Wouldn't the test group and the control group of mice both be equally affected by the noise? If the thing being tested for really was making a difference, shouldn't that difference still be visible? This sounds an awful lot like: We went into the experiment knowing what results we wanted, and we twiddled knobs and kept discarding "bad" data under the thinnest pretenses, until we finally got them. And that's not how you do science.

    When it comes to the researchers whose polymer was being degraded by UV photons from normal daylight... I'm sorry, they just don't sound very smart. I have to wonder if their situation would have improved if they had installed this monitoring system. What would it have told them? "Your experiment is occurring at room temperature, earth gravity, normal daylight, air of terrestrial composition, yadda yadda." Are the salesmen suggesting that these bumbling scientists would have looked at all this "data", slapped their foreheads and yelled: BY GOD, WE JUST LEARNED THAT OUR EXPERIMENT IS OCCURRING IN NORMAL DAYLIGHT!