Automating Research with SwiftSCIENCE
SwiftSCIENCE is working with Dr. Vince Setola’s lab at West Virginia University as partners to automate how they quantify behaviors. Dr. Setola said “I value working with partners who are scientists, who understand so much about basic science, medicine, and the treatment of disease, and an engineer with signal processing expertise. We are getting really specific about defining what a specific behavior is and how we count it.”
SwiftSCIENCE consults with animal researchers to automate their work. This process usually takes about eight weeks. The following graphic shows the steps that we take in our artificial intelligence (AI) for animal research.
SwiftSCIENCE collaborated with the lab for manual behavior detection and labeling. SwiftSCIENCE put the cameras onto cages, captured 15 minutes per day looking for a single behavior in the videos, and then WVU staff manually documented and labeled behaviors for the first study. SwiftSCIENCE provides the fastest way available to label data, saving investigators a lot of time.
SwiftSCIENCE then manually documented behavior data and automated the process by training a model to distinguish and document the behaviors automatically. Once complete, SwiftSENSE’s system was able to passively compute and report on copious amounts of data over a few days. SwiftSENSE eliminated $108,000 and 4,320 hours of human labor and improved the behavioral data evaluation compared to previous human methods. Our toolset can significantly reduce the time and budget allotted to manual processes throughout the research industry, while improving accuracy. The four behaviors that the machine learning identified with a 95% or greater accuracy include:
Body Lick Grooming
Hands to Face Grooming
Dr. Setola says “Now that we know some behaviors that are associated with meth withdrawal, we have some ideas that could help cure these withdrawal symptoms. If they work in mice, they may work in humans.” Getting these data points in the hands of people who understand signal to noise, understanding mouse behavior, training set versus test data, is where the magic happens.
Eventually, they envision having a rack of mice, with five mice in each cage, and each row has a different treatment to test. By using the SwiftSENSE solution, they will be able to accelerate their research by automating data collection, evaluation, and testing multiple treatments at once.
SwiftSENSE will allow investigators to delegate scoring of the behaviors, allowing the researcher to do more research, conduct more studies, and input more data to analyze efficiently. The more data that you feed into the system, the clearer the results get. Comparing data gathering and recognition between computers and humans is vastly different. As a human, if you are not looking for a particular behavior or only scoring one other behavior, you may miss the other behavior that was clinically more significant. This is especially true if staff are watching the video at two times the speed over prolonged periods. For example, imagine within 50 seconds of a grooming behavior video, you have a wet dog shake behavior thrown in. Depending on the speed you are watching it at, most humans could easily miss the wet dog shake behavior while looking for and counting grooming behaviors. As scientists, you have a hypothesis, and it may influence what you witness while manually reviewing many hours of video data. Computers are unbiased, do not get fatigued or distracted, and will just identify the pattern, including something that humans may not have caught. Evolving this assessment process with technology allows the drug discovery timeline to be completed faster.
Dr. Setola shares, “We’ve had several very exciting discoveries that we’ve collected in collaboration with SwiftSCIENCE, including some we’ve already divulged to National Institute on Drug Abuse (NIDA). They found it very interesting. They are very concerned about the alarming escalation of meth use.”
He also believes IT departments at universities need to be more open to working with outside groups and companies if they want to be true partners in discovery and innovation. He advocates for IT departments to facilitate collaborations with companies like SwiftSCIENCE and to allow tools, such as SwiftSENSE, to be allowed on the university network more readily.
Dr. Setola and the team hope to publish some of their initial results by the end of this year, sharing behaviors that they can target to help meth users in withdrawal. Then their research will move on to looking at withdrawal behaviors seen with other psychostimulants. After collecting all the information about behaviors, they will move on to try treatments that can ameliorate some of them. They have a hypothesis about receptors in the brain that can be used to treat the problem.
“We’re validating the SwiftSENSE solution through our work but hope that the platform can be used by other researchers after validation to make other scientific discoveries, move things forward, and benefit all of mankind,” said Dr. Setola.
SwiftSCIENCE is thrilled to be partnering with Dr. Setola and his team for this research. We look forward to sharing more information with you after his research is published.