Fixing the Lab Reproducibility Crisis with Augmentation, not Automation
BioBright strives to augment human ability to do science in the lab with voice assistance that recognizes biomedical research terms and the use of augmented reality tools.
A recent Nature survey of 1,576 researchers found that more than 70 percent of respondents have tried and failed to reproduce another scientist's experiments. More than half have failed to reproduce their own. This “reproducibility crisis” was revealed to be especially acute in biological and medical research.
Most of the Nature survey respondents agreed that the solution to the crisis is improved documentation and standardization of protocols. Yet given time pressures, few were willing to put in an estimated 30 percent more time required for comprehensive documentation.
“Current laboratory tools are not conducive to reproducing experiments,” says Fracchia, a bioelectrical engineer who is currently on leave from the MIT Media Lab. “Most tools don’t connect to networks or have outdated connection mechanisms and proprietary formats. All that stands in the way of doing longitudinal, data driven discovery.”
To ease the documentation process, Cambridge, Mass. based BioBright offers a suite of “smart lab” software, hardware, and services. The company built a voice assistant called Darwin tailored to recognize biomedical research terms, and software that automatically collects data from laboratory equipment. Information is recorded, aggregated, and analyzed in the cloud where other scientists can access this integrated record in order to duplicate the experiment.
“Our goal is to augment the human ability to do science in the lab with voice assistance and augmented reality tools,” says Fracchia. “With BioBright, scientists can spend their time uncovering the root causes of success or failure while staying in the experiment. They don’t have to stop every 15 minutes to record information. Our system allows the lab notebook to write itself so scientists can do what they’re supposed to do -- analyze the data.”
BioBright is targeting biomedical researchers in academia, biotech, pharma, and even healthcare, and the tools could eventually expand to other industries such as food research. The company was recently chosen to be among the first six companies in the new MIT Startup Exchange 25 (STEX25) program. This more focused version of the successful MIT STEX program is designed to facilitate industry interaction with the 25 most promising MIT-based startups.
Augmentation over Automation
Some have argued that the reproducibility crisis can be solved with laboratory automation -- replacing human-driven lab processes with robots and other devices. Fracchia himself helped launch an earlier MIT-based startup named Gingko BioWorks that built the first automation tools for synthetic biology.
The experience taught him that while automation is often valuable, it does not necessarily improve reproducibility. “Automation is excellent at doing workflows and protocols 1,000 times over, but it’s very brittle, and has no ability to adapt,” says Fracchia. “Research is for the most part an exploration. You’re adapting and changing parameters on the fly. BioBright’s augmentation tools keep humans at the center of the loop, not computers.”
Fracchia goes on to note that the automation world is mostly driven by computer scientists “who don’t know the intimate problems of the lab.” BioBright inverted that. “We are biologists with intimate knowledge of biomedical workflow who learned computer science and electronic engineering.”
According to Fracchia, most reproducibility problems stem from translating workflow between researchers. “Automation won’t help with that,” he says. “BioBright’s human augmentation, however, allows the scientist to say ‘Darwin, show me the average temperature that I’ve used for the last three months, or show me how Mike did it last week.”
BioBright helps reduce the often significant time lapse between data generation and collection, says Fracchia. “Researchers’ hands are often busy and gloved, so they must put off the time when they record information, sometimes even to the end of the day,” he says. “A lot of information gets lost this way. Small deviations from the protocol are often not recorded. Maybe you will note on a Post-It that a sample was a little more viscous than usual, or if you’re really disciplined, write it in a notebook, but there it stays. It’s difficult to look at the information in context.”
BioBright helps solve a related challenge with laboratory documentation: the vast range of time scales. “In biological research you’re ranging from picoseconds to hours, days, or even months,” says Fracchia. “Humans are good at collecting information in the minute range, but have difficulty in other ranges, especially picoseconds. BioBright pervasively collects information across different time scales and centralizes it in one place.”
Inside BioBright – from Custom Sensors to Voice Control
The sensors and cameras available with BioBright vary depending on the specific environment, but the underlying architecture remains the same. BioBright is a cloud-based platform that uses a modest onsite computer – currently a Raspberry Pi board – to act as an Internet of Things hub. The local device aggregates information from the lab sensor network and mediates with the cloud service using end to end encryption.
One of BioBright’s key innovations is a “hot folder” that interfaces directly with lab equipment. “We automatically grab the data as it’s generated and centralize it,” says Fracchia. “The system is built to be extensible and include new data formats based on customer’s needs, which allows us to extract metadata that is directly relevant to our customers’ workflows.
BioBright offers camera systems that record in both visible and infrared light. “You can place these cameras around your lab, or over benches or specific stations,” says Fracchia.
The company has also developed tiny sensors designed for biological research that can fit into standard sized vessels or tubes. “We developed the first temperature sensor that fits in an Eppendorf tube,” says Fracchia. “Our wireless sensor lets you easily record the temperature of your samples across an experiment.”
BioBright’s Darwin voice assistant enables researchers to issue voice notes instead of stopping to record information manually. “You add a little microphone to your lapel so you can interact with BioBright, and leave voice notes,” says Fracchia. “You can also give orders like telling the camera to record an image.”
The natural language AI system also works in reverse, letting you ask the computer for information or to correlate data. You can even set up the system to volunteer advice based on sensor input and historical data.
“BioBright is bidirectional, sending longitudinal information back to the scientist,” says Fracchia. “For example, we can warn scientists that they’re conducting a test at the wrong temperature. Even if they ignore the warning, they can analyze the difference between what actually happened and what was supposed to happen. At an early stage, we can tell you that your protocol is unlikely to succeed, which is tremendously important in pharma and biotech where it can take weeks to get results.”
BioBright is built on modular components that the company assembles for customized services sold to large industries. Some components, however, will be sold as standalone products.
Although most of BioBright’s technology is proprietary, Fracchia is a proponent of open standards and interoperability, which are often lacking in the biomedical field. “We use interoperable data formats so you’re not trapped in the ecosystem,” he says. “BioBright solves a problem that is common across a number of industries: scattered data in different formats.”
The company is now working on integrating sensor networks, wearable sensors, and augmented reality head mounts into the system. Eventually, Fracchia envisions something like an Iron Man suit for biomedical researchers. “We want BioBright to fit the workflow of the scientist like a glove.”
Impact before Income
BioBright draws extensively on MIT Media Lab’s innovative pervasive computing technology. Fracchia mentions the Media Lab’s Principal Research Scientist Shuguang Zhang as being especially helpful in launching the company. Other MIT institutions have also played a big role.
“I cannot say enough good things about the Venture Mentoring Service, which has been instrumental in getting BioBright to where it is today,” says Fracchia. “And MIT’s Technology Licensing Office helped us look at innovation in a refreshing way. They really understand MIT’s motto: Impact before income.”
That motto has steered BioBright away from venture capital for the time being. After raising a modest angel round, the company has been sustained entirely by customer contracts, which Fracchia says is quite unusual. “A four-year return rate was not the best match for us,” he says. “We don’t have to make investors happy or report to a board that is driven around valuation. Instead, we can focus on partnerships and analyzing companies’ reproducibility problems. We want to solve problems, not just sell you a product and go away.”
Fracchia says his team is “very honored and humbled” to be named to MIT’s STEX25 program. “STEX and the ILP have let us grow and better understand our customers,” he says. “The point of STEX25 is to link up transformative technologies with companies that are creating real value.”
STEX25 is a startup accelerator focused on fostering collaboration between MIT-connected startups and member companies of MIT’s Industrial Liaison Program (ILP). STEX25 is managed by MIT Startup Exchange, and its parent, the ILP. The ILP is a key player in making industrial connections for MIT, with over 220 of the world’s leading companies using their ILP memberships to advance research agendas at MIT.
- Winning Back Unresponsive Customers with Artificial Intelligence and Machine Learning April 24, 2017
- Small Satellites with Huge Capabilities
Propelled by Ion Engines March 28, 2017
- New Startup, Gamalon, Invents And Commercializes Revolutionary Machine Learning Technology February 23, 2017
- Apps for Operators on the Factory Floor December 5, 2016
- Gradiant Replicates Rain Cycle
for Fresh and Recylable Water November 30, 2016
- Last Mile Delivery and Logistics Made Smarter Through Wise Systems November 30, 2016