When I realized that $28 billion is wasted in life science research every year, I wanted to figure out the scope of the problem and come up with a solution.
PLOS Biology published data on the $28 billion lost in preclinical research back in June 2015. It raised a vital question: What is the life science community actually doing now to prevent this colossal amount of wasted time, money, effort and inventory?
To put this in perspective, every year the biotech and pharmaceutical industries collectively spend more money on drug research and development than the combined revenues of the U.S. film industry, the music industry and publishing industry. And half of that estimated $56 billion total is wasted on irreproducible research.
Indeed, this is an expensive, horrendous problem facing life science today. I spoke about it at a recent daylong forum “Accelerating Science Through Innovation” at Lab Central in Cambridge, MA.
Track the "Unknown Unknowns"
Science is complex, and it isn’t easy to devise a solution when you’re looking at “unknown unknowns” – the vast array of causes of failed experiments due to myriad factors like environmental variations, process errors and instrumentation faults – and enable scientists to avoid the failures. In other words, you don’t know what you don’t know. Where do you start? Where do you look?
This issue of tracking the “unknown unknowns” initially came to my attention when I ran into problems with scaling up the manufacture of glucose strips at my first startup, AgaMatrix. By measuring many different variables in the factory (e.g. via sensors and other data sources) in the factory where the strips were made, we were able to correlate fluctuations in many different variables (such as temperature, humidity, machine maintenance) that were rendering whole batches of the materials unusable. Once we were able to see these correlations, we were able to essentially predict production issues well in advance and take preventative action very successfully.
Some real-life examples of irreproducible experiments include things like:
- The noise and vibrations of nearby construction (or the cleaning staff leaving the lights on), stressing lab mice and affecting their response to certain drugs or disrupting their breeding ability.
- Doors to temperature-controlled storage being left ajar all night and destroying priceless inventory.
- The HVAC system directly blowing air on critical equipment and distorting results.
At my current startup, we have developed a suite of sensors and artificial intelligence tools to measure everything from temperature to light to air moisture levels to machine usage. In this way, we can enable companies to more quickly identify the wide range of variables that could be influencing their experimental processes and outcomes.
“Silly Things That Went Wrong”
Problem-solving wasn’t about fancy predictions in cases like these. The limitation was in the gaps of data – there just wasn’t enough data being collected and automatically analyzed to spot time-dependent anomalies. As a result, it could take weeks or months to debug what went wrong. But once you finally figured out what did go wrong, it often took only minutes to fix.
Half of the scientific research being done today is not reproducible due to factors unrelated to the scientific theory being investigated – often the unknown unknowns noted above fall into the category of “silly things that went wrong” and not due to any fundamental lack of scientific understanding of the science. Scientists are constantly confounded by results from the same experiments that vary on different days. They must repeat them at great expense in time and money to try to achieve the same result.
Make It Easy to Measure Everything
What does a startup do when temperature and humidity-controlled chambers are exorbitantly priced and not feasible to purchase? How does an animal testing lab with a limited budget move forward from using all paper records for room environmental tracking so that it can track the conditions in real-time? How does a facility with shares lab space and hundreds of assets track all its cold storage equipment? How does it connect the data? It needs real-time monitoring of temperature as well as stats on compressor cycles, how frequently a door is being opened, and if the door is being left open, all of which allows the staff to also be proactive about preventing issues and ensuring a stable facility platform for doing science.
I wanted to develop technologies and products that could measure what we don’t know about failed experiments so that we can prevent this colossal waste of time, money, and resources in scientific research. Scientists need peace of mind to save time and improve the accuracy of their studies. They need this assurance 24/7 with an AI platform that can monitor instruments and processes for any change in the myriad variables that could affect their work, including the environment, instrumentation, reagent storage, etc. They need immediate alerts and real-time data if a problem arises.
Sensors coupled with AI models can be used to measure everything from temperature to light to air moisture levels to machine usage. In this way, companies can more quickly identify the wide range of variables that could be influencing their experimental processes and outcomes.
Scientists need products and technologies to improve the accuracy of their studies and give them peace of mind. They need this assurance 24/7 to monitor instruments and processes for any change in the myriad variables that could affect their work, including the environment, instrumentation maintenance, reagent storage, calibration, and so on. They need immediate alerts and real-time data if a problem arises (or even better, if it is about to arise).
Ultimately, our view is that much of science is slowed down by experimental hiccups that can often be traced to the unknown unknowns. The solutions? By making it easy to “measure everything”, we aim to bring clarity and certainty to experimental results and hope to empower scientists to focus on their science and let technology worry about keeping a watchful eye on what could go wrong.
Sridhar Iyengar, Ph.D., is CEO of Elemental Machines, a Cambridge, MA-based firm helping scientists invent faster by improving experimental reproducibility.
References
- PLOS Biology, 6-9-15, The Economics of Reproducibility in Preclinical Research, https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.1002165
- Statista: Box office revenue in North America from 1980 to 2019, https://www.statista.com/statistics/187069/north-american-box-office-gross-revenue-since-1980
- Recording Industry Association of America (RIAA), 2-28-19: RIAA Releases 2018 Year-End Music Industry Revenue Report, https://www.riaa.com/riaa-releases-2018-year-end-music-industry-revenue-report/
- Forbes, 7-2-18, Estimated U.S. Book Publisher Revenue Was North Of $26 Billion In 2017, https://www.forbes.com/sites/adamrowe1/2018/07/22/estimated-u-s-book-publisher-revenue-was-north-of-26-billion-in-2017/#64232ace3196