How did your PhD inspire you to work in the area of cancer genetics?
I worked in the lab of Tony Wynshaw-Boris at UC, San Diego and we were investigating the ATM gene, which turned out to have an unexpected role in cancer. Interestingly, humans who have just one mutated copy of the gene, have a 15-fold increased risk for getting cancers.
In Tony’s lab we had made genetic knockout mice to mimic that genetic condition—having one mutated copy of ATM—then, we investigated what effect the mutation had on cancer risk. I found that these mice didn’t get more cancers, but the cancers they got were more aggressive compared to the mice without the ATM mutation. I had to seek out pathologists to help me characterize these tumors, and that taught me that pathology is a really challenging discipline.
What else did you learn while collaborating with pathologists?
I was blown away by the nuances they could see both with the naked eye, sometimes just holding a glass slide up to the light, and through the microscope. I’d be looking through another scope beside them, and it was clear how much education and experience looking at cancer cells played into their daily work.
It was also apparent how busy they were—these were clinical pathologists and I often couldn’t get time with them. They were so overwhelmed with work looking at patient samples that there was not always enough time in the day to help with research projects.
And finally, I saw that there were times when two pathologists needed to review the sample to get the diagnosis right. I’d sometimes have to get them on the phone together to come to an agreement. If a pathologist misses one tiny, but critical element, it can change the actual diagnosis.
How did your experiences after the PhD continue to shape your view of pathology and patient care? I first went into strategy consulting for the pharmaceutical industry, advising on drug development and clinical trials. Eventually, I moved to Cardinal Health’s Innovation Group, thinking about how we could better serve their clients—about 80% of the hospitals and cancer centers in the US—through technology.
We were looking at ways to help community oncology clinics run more efficiently. This experience got me even closer to understanding the pain points of oncologists treating cancer patients. I met a lot of oncologists who explained how hard it was to get patients on the right trial and we had a lot of conversations about precision medicine—and what it would mean to specifically target a therapy to a patient’s particular cancer. During my PhD, in the early 2000’s, it always felt ‘wrong’ that physicians treated patients based on the tissue that their cancer was in—say, lung or colon—rather than basing treatment on the mutations or molecular pathway mechanism of the cancer.
It was super exciting to see people recognizing that mechanism was important and that every cancer is unique. It drove home how important an accurate diagnosis is to find the best precision medicine. Why did you decide to co-found Deep Lens as a startup digital pathology company? When Dave Billiter approached me with the idea of starting Deep Lens, the large pharmaceutical companies were just starting to focus on pathologists because they are the ones it starts with—making the specific cancer diagnosis for a patient. Because today’s options for cancer treatment are so complex, and based on the molecular signatures of each cancer, that means that pathologists are also the ones who might miss critical diagnostic assays or biomarkers needed to inform the best treatment decision.
We wanted to make the pathologist’s workflow easier, more efficient, and more precise. But we were also thinking more broadly about how we could enhance this VIPER digital pathology platform, too. We wanted to see if we could solve this bigger problem of getting patients to the right clinical trials based on early information and the correct diagnosis. How will VIPER accomplish that? We searched for data scientists who could design artificial intelligence software to enhance VIPER. We found a group of deep learning experts in our own backyard at Dayton University. Deep learning is a type of artificial intelligence based on the way the human brain works to build visual representations of our world.
Importantly, they were using different algorithms than what other digital pathology groups were using. Their algorithms were first developed to do military-style aerial surveillance of images. We asked them to apply those algorithms to pathology images. They found that their AI was significantly more accurate at identifying cancer cells and other cell features. The value of using these new math models is that they can be just as accurate, if not more accurate, using less data than other types of AI. That’s significant, because data is more valuable than any currency in the world right now and it is scarce. Any AI has to be “trained” first on a sample set of known images, with known diagnoses, known treatment paths, and known patient outcomes attached. But for any given cancer subtype, we might only have access to a 1,000 known samples at best, and not the tens of thousands that other AI programs need for training.
If we can train VIPER’s AI on fewer samples, but still be able to accurately predict a diagnosis, a treatment option, or even a prognosis, that’s invaluable. VIPER’s AI is also designed to handle the mundane tasks of a pathologist’s day, such as counting cells or calculating how many cells are dividing. We’re creating VIPER in a way that will be open, so that as new AI technology becomes available, say for identifying a specific tumor type, it can be plugged right into the platform. Why are you so passionate about connecting physicians, caregivers, and patients to the best clinical trials and treatments available to them? Just after graduate school, my aunt was dying from ovarian cancer. As someone with a PhD, I was the only person that my mom and my uncle could turn to for answers about my aunt’s treatment. My uncle lives in a rural area where there isn’t a major cancer center. After my aunt’s standard therapies failed, I helped my uncle navigate the clinical trials that she could join. It opened my eyes to how difficult that would be for someone without an advanced biology or medicine background.
Many people aren’t fortunate enough to live in a big city and they aren’t always getting access to optimal treatment. How will Deep Lens make a difference? Deep Lens is uniquely bringing the pathologists into the realm of clinical trial support. By making sure that the pathologist, oncologist, and care team are connected as soon as possible, when the diagnosis is made, we can ensure that they are getting information about treatment decisions as early as possible. At that initial diagnosis stage, VIPER will pull together all of the patient’s available data, cross-reference it to ongoing clinical trials and then suggest trials that the patient might be eligible to participate in. At the same time, as more pathologists use VIPER globally, we’ll be able to see data trends that can help pharmaceutical companies identify where and how to set up clinical trial sites for the best chances of enrolling qualified patients. Ultimately, we hope this leads to both patients finding their best choices for treatment and more successfully completed, faster clinical trials.