Animals used in research have been monitored in much the same way for at least the past 60 years. Behavioral and physiological study endpoints typically are observed only at a few discrete timepoints throughout the study and are often subjectively scored.
With that approach, “Reproducibility is a challenge, and it leads to translatability issues,” Michael Ellis, PhD, senior director, research and development at The Jackson Laboratory (JAX), tells GEN. “To effectively understand biology, and to understand the solutions needed to get a compound to the clinic, we need to address reproducibility and translatability.”
Continuous monitoring of mice in their home cages is the solution, but manual video review is time-consuming and tedious. A better option uses artificial intelligence (AI) to identify relevant details in the study animals as individuals or as groups. This approach allows researchers to concentrate only on the details that matter.
Envision™, developed by JAX, is a validated, cloud-based solution that uses computer vision, a specialized field of AI, to monitor the vivarium 24/7 for relevant behaviors and physiological measures, as well as conditions, like food and water levels, that affect the animals’ welfare.
JAX recently partnered with Allentown, a provider of vivarium solutions, to launch a fully integrated AI-driven mouse home-cage monitoring solution. Combining Envision with Allentown’s new Discovery™ IVC rack allows scientists to better capture the behavioral and physiological data needed to advance drug development.
AI-enabled continuous monitoring
JAX’s Envision platform integrates validated AI models that have been trained to recognize and interpret certain activities and behaviors (like seizures, respiration rates, distance travelled, and activity levels), relieving scientists of the tedious task of manually reviewing hours of video. Consequently, studies can rely upon standardized analysis of continuous data capture, which thus improves study reproducibility and reduces resource use.
Each Envision model is trained on data that accounts for variations in coat color, bedding, and enrichment. Envision tracks individual mice, recording details of their behavior and physiology, and provides alerts for food and water levels to help ensure proper care.
As a cloud-based application, Envision data is accessible 24/7 anywhere an Internet connection is available. Consequently, researchers can study individual or group behaviors and annotate studies without entering the vivarium. That’s convenient for researchers and less stressful for mice.
Changes detected early
One of the quirks of animal research is that experiments typically have been conducted during the day, “but mice are nocturnal,” Ellis points out. “We’re conducting experiments during their natural sleep cycle, which is stressful for the mice and suboptimal for the science.
“For context, imagine being awakened in the middle of the night, grabbed by a giant, pulled out of your home, and experimented on,” he says. “With Envision, however, we’re just watching.” Stress on the animals, therefore, is minimized and the data are more translationally relevant.
The data that emerge from continuous versus periodic monitoring can be significantly different and can affect drug development programs.
In one study, for example, mice carrying a genetic modification with a phenotype that resembles amyotrophic lateral sclerosis (ALS) began to deviate from normal activity at about seven weeks of age and continued to worsen. Symptoms were particularly evident “during one three-hour period at the end of the dark cycle, when mice are about to go to sleep. The worst phenotypes come out then,” Ellis says, but that isn’t when most studies are conducted.
“Using normal testing approaches, researchers wouldn’t detect that behavioral change until 14 weeks,” he says, “but we detected it at seven weeks. For drug developers, that can significantly reduce time and costs.”
High data volume
“The challenge in building computer vision models is the large volume of high-quality data necessary to build high-quality models,” Ellis says. In addition to JAX’s pre-built behavioral models, researchers can use the Envision Developer Environment to customize Envision and track the specific details they want to measure.
“The analogy is that Envision is like an operating system, and we’ve built respiration, seizure, and activity monitoring apps, and an app store,” Ellis says. The ability of researchers to customize the behavioral and physiologic features they monitor will, he suggests, help Envision evolve in concert with study needs.
Continuous monitoring also creates a vast body of potentially valuable data, and that makes each animal more valuable than ever. Previously, researchers may have recorded roughly a half hour of data from each subject per day. The additional data generated through continuous video-based monitoring not only informs the current study, but also allows researchers to go back and explore other details that may not appear relevant until later in the drug development process.
This capability not only saves time and money by potentially eliminating the need to conduct an additional mouse study, but also allows scientists to consider that data in context, because the follow-on study isn’t conducted with mice like those in the first study, but with the original mice.
“Also, Envision is just as capable of identifying animal welfare issues as scientific issues,” Ellis adds. “We can see when a mouse isn’t doing well and can step in.” In addition to real-time or historical viewing, researchers can set up notifications for when food or water levels are low or when cages need changing.
Future directions
Envision was designed to address the needs of preclinical researchers, Ellis notes. That ranges from principal investigators in academia to pharmaceutical industry biologists, veterinarians, pharmacologists, and data scientists.
He says biopharmaceutical organizations already are using this solution to identify seizures for epilepsy studies. “Seizures don’t happen often in a study—maybe three to five times per week—so they’re hard to find by simply reviewing video.” By training an Envision model to recognize seizures, however, tedious manual monitoring can be eliminated, and scientists can focus on the events that matter.
AI-enabled animal monitoring has the potential to significantly change the way animal research is conducted, making it less invasive and less disruptive for the animals, and more productive for the scientists gathering and interpreting data.
“Thus far we have only addressed the tip of the iceberg regarding what this tool can measure,” Ellis says. “We’re focusing on working with other organizations to build out more digital measures.”
One of those is the Digital In Vivo Alliance. This pre-competitive alliance of leading biopharmaceutical companies, which includes JAX and Allentown, is developing and clinically validating behavioral and physiological measures for a variety of disease models.
It’s fair to say that how animals are monitored during scientific studies is about to change dramatically. And, when that happens, each mouse may contribute more to science than ever before.
