Getting Ph.D.s out of the mouse labs
A startup aims to take humans out of labs that test drugs on mice Can software replace “a curious person observing firsthand”?
Much of the groundwork for important new drugs starts with tests conducted on lab mice, of which there are an estimated 100 million around the world. The Ph.D.s and technicians who observe them often spend their time on rote tasks that a Silicon Valley startup called Vium wants to automate.
Co-founders Timothy Robertson, a physicist, and Joe Betts-Lacroix, an inventor who has sold inventions to Google, started Vium in 2013 with a mission to update mouse labs using cheap sensors, deep learning software, and virtually no humans. “We derived more data and better-quality data without exposing the mice to the stress of human contact,” says Robertson, who is also chief executive officer.
Vium’s 18,000-square-foot lab in Milpitas, Calif., is dimly lit in red light that mice can’t detect, and almost totally silent. The only sound you hear is the faint rustle of mice moving about in their rows of cages. Each cage is illuminated by LED spotlights—white for daytime and infrared for night. In traditional labs, humans regularly perform tests on the mice and record periodic observations. In the Vium lab, an array of sensors, including tiny, highdefinition cameras wired above the cages, collect physiological data and track behavior. That data are then fed into an algorithm that delivers real-time readings on the animals’ well-being. Reports show behavior patterns such as lethargy or clumsy walking, but humans will still clean cages.
Over the past two years, Vium has raised almost $30 million from venture capital firms, including Peter Thiel’s Founders Fund. The idea is that an entirely automated system can speed up testing and cut the costs of bringing a drug to market. Investors are hoping Vium can poach pharmaceutical clients from so-called contract research organizations, or CROs, mouse-for-hire labs that earned about $6 billion last year.
Vium executives argue that many CROs have cumbersome procedures and rigid reporting practices that provide data only at the end of a scientific test. Vium intends to make setup as easy as reserving server space on Amazon Web Services. Customers simply provide their credit card, fill in a few web forms, and then mail or deliver their drug compounds to
Vium’s lab. Once testing begins, customers can view data on their phones or laptops almost immediately.
Kenneth Kaitin, director of the Tufts Center for the Study of Drug Development, is skeptical about Vium’s ability to lower costs. “The barriers to faster drug development aren’t in preclinical,” he says. “All of the enormous inefficiencies—and costs—are in trials with human subjects.” Yet Vium investor (and former Pfizer CEO) Jeffrey Kindler says he’s betting a smarter preclinical stage is valuable. “The earlier in the drug development process that you predict how the drug will perform in humans, the better,” he says. “It allows you to fail quickly, terminate programs, or redesign them.” He credits Vium for taking a fresh look at the mouse research industry, especially at problems with human intervention. “What these guys have done is take the subjectivity out of animal models.”
Most of the work Vium aims to automate is done by postdoctoral students such as Megan Tipps, a University of Minnesota at Twin Cities researcher who conducts addiction studies. Tipps says mice generally adapt to being handled by humans and questions an algorithm’s ability to replace the human element. “I don’t know that there will ever be a replacement for a well-trained, curious person observing firsthand.”
So far, clients include AbbVie, a $23 billion pharmaceutical company developing a hepatitis drug, a Harvard researcher testing an anti-aging treatment, and a drug-discovery startup run by former Princeton researcher Ethan Perlstein. “We had been working with a traditional CRO, where the technicians were physically pulling the animals out once a day to weigh them,” says Perlstein. “With Vium, you have a scale built right into the cage and other features to monitor them 24/7. Perfect!” Perlstein believes removing the human factor “goes a long way to making studies reproducible.”
The proof will come as more clients use Vium’s system and publish their data.
Instead of humans, Vium uses tiny sensors and high-definition cameras to observe lab mice