Big buzzwords in computing and biotechnology fly fast and loose in the headlines about research innovation. A quick glossary spells out the science behind the hype—and shows how Penn Medicine is driving discovery forward with these technologies.
By Frank Otto
Illustrations by Graham Perry
Artificial Intelligence (AI)
noun A form of adaptable computer programming that is seen to approach or resemble the abilities of human intelligence.
In recent years, artificial intelligence (AI) has become one of the most potent tools at the health care industry’s disposal. It’s used to identify cancerous tumors through pathology images, discover patterns in patient charts to head off potentially deadly infections, and comb social media to predict who might be at risk for depression.
As impressive as AI is, the barrier to entry for its use is pretty high. First, AI software is expensive, which often makes it unreachable to even mid-level health systems, much less individual researchers who have ideas they want to tinker around with. And second, the tools that exist often require advanced technical expertise.
“The problem with AI tools is that ‘AI people’ build them, so they’re usually only usable by those with high levels of training,” said Jason Moore, PhD, one of Penn Medicine’s leading AI gurus and the head of the Institute for Biomedical Informatics.
Moore said that fellow researchers, including many at Penn Medicine, have asked him about the best AI software and tools for beginners. His answer has always been, “There isn’t any.”
Now his answer has changed.
In May, after three years of development, Moore and his team officially launched Penn AI, a completely free AI data analysis tool that was designed to be usable by anyone from the beginner—say, a high school student looking to gain insight on their baseball team—to the seasoned data veteran. An automated machine learning system, Penn AI uses an artificial intelligence engine to work out analyses of datasets with minimal human input using different variables and methods. Users can upload their custom datasets—anything from a sports team’s statistics to a spreadsheet with cancer treatment data—adjust the parameters with guidance from the software’s built-in help information, and let the tool do the rest.
“The feedback has been overwhelmingly positive,” Moore said of Penn AI since its launch. “In addition to comments about its ease of use, we have received a number of requests for new features that we plan to add over the next year.”
Moore and his team are also working with Penn Medicine Information Services to integrate Penn AI with a different project, PennTURBO, which works to make better connections between relevant clinical data and research efforts.
Moore envisions a (possibly near-) future where AI data analysis is as common as a doctor checking a patient chart. For a clinician-researcher, some of the most time-consuming parts of discovery, from hypothesis-generation to analyzing results, could now be run automatically while on their normal rounds—including complex analyses that were never in their toolbox before.
“I want this to be self-service, clinical AI,” Moore said. “I believe that this tool can make it so that it will soon be routine for a doctor to say, ‘I want to look at the associations between sex, age, smoking and different diseases,’ and then have this tool answer their questions in the time it takes for them to see a few patients.”
Penn AI is available online as open source software. Visit pennai.org to access the source code.
noun (plural) Non-human, AI-powered tools that mimic selected human activities without supervision.
Penny knows the complex pill regimens of chemotherapy backwards and forwards. She can tell you when to take what pill, which combination is coming next, and even help manage the side effects that often crop up.
“People like Penny,” said Lawrence Shulman, MD, a professor of Medicine and the director of Penn Medicine’s Center for Global Cancer Medicine.
It’s true. The 10 patients who texted with Penny all gave her extremely high marks. In addition to their clinical interactions, patients shared news of their day or texted emojis to celebrate sports victories.
Penny isn’t a superstar staffer. She’s a chatbot, an artificial intelligence-run texting platform developed and piloted through Penn Medicine’s Abramson Cancer Center and the Center for Health Care Innovation in collaboration with New York-based startup Patient.ly.
“It’s a personal connection even though it’s a machine. I feel responsible to respond,” said Teresa Sweet, the first patient to begin texting with Penny, back on Christmas in 2017. She feels so connected she makes a point of reassuring the bot if she needs to delay her medications, texting in advance, for example, “I’ll be out for the evening and I won’t be able to take anything until 8.” “I don’t want the thing to worry about me,” she said.
Penny’s job was to help patients with neuroendocrine cancers take chemotherapy pills at home. Over the last decade, there has been an increasing shift toward outpatient therapy. The thought, Shulman explained, is that keeping a patient home makes them more comfortable and assists in their healing. However, the at-home regimen can be difficult to follow. Many times, it involves varying combinations of eight to 10 pills over 10-day periods. It can get confusing fast, and there are other precautions to remember, such as washing hands after handling the pills because they are poisonous. Penny provides regimen reminders and interacts with patients when they’re experiencing side effects to provide tips on overcoming them.
Penny was developed for the two most difficult oral chemotherapy regimens in neuroendocrine cancer treatment, CAP/TEM (capecitabine and temozolomide) and capecitabine, specifically because of how challenging they are.
“If we can figure out a bot for these two drugs, we can figure it out for any of them,” said Penny team member Christine Cambareri, PharmD, an outpatient clinical oncology pharmacy specialist.
Because they engage with patients so often—and are built to automatically keep records of these interactions—bots are providing clinicians with more clues of the hiccups can occur in at-home regimens.
“We knew that there were some deficiencies in the home treatment, but we truly didn’t know where they were occurring,” said Beth Mooney, CRNP, MSN, a certified registered nurse practitioner in Gastrointestinal Medicine. “Now we can see where issues are.”
Penny shows promise with several other areas beyond cancer, such as OB/GYN and primary care—all areas where future Penny pilots are under consideration. And, according David Asch, MD, MBA’89, the executive director of the Center for Health Care Innovation, bots may portend a future path to health care efficiency: Writing this spring in the New England Journal of Medicine, Asch and colleagues described the potential use of bots in health care as “facilitated self-service.” In the way that ATMs and cashless banking apps have reduced the need for most interactions with bank tellers, health care bots could be used extensively to free up clinician time.
“Clinical pathways for common medical conditions aim to make care algorithmic,” they wrote, “so it isn’t science fiction to suggest that hypertension could be managed using a bot, with a nurse available for second-line support and a primary care physician serving as the third line.”
noun A family of DNA sequences with applications as a genetic engineering tool that uses a sequence of DNA and its associated protein to edit the base pairs of a gene (stands for Clustered Regularly Interspaced Short Palindromic Repeats)
“You take a shot that serves you for the rest of your life. Once and done.”
It sounds like Kiran Musunuru, MD, PhD, MPH, ML, is describing how polio and smallpox have been stopped in their tracks. But what he’s actually talking about is something entirely different: High cholesterol.
“With a single shot, we can permanently turn off cholesterol genes in the liver,” said Musunuru, an associate professor of Cardiovascular Medicine and Genetics. “It’s like taking a statin every day for the rest of your life, but you just have to do this once.”
Such one-and-done therapies are already emerging thanks to newer gene therapies that add a healthy working gene to cells that lack it—with several such therapies developed at Penn. Now CRISPR is widely seen as the tool to exponentially increase the use of these therapies through gene editing. Nimble and inexpensive, instead of just adding a healthy gene into a hole, CRISPR can replace a gene. It uses a guide RNA and an attached protein to seek out and replace matching DNA components. Put simply, the tool cuts and pastes good genes onto bad ones. If a scientist can find the bad genes that lead to disease, CRISPR is a path toward “turning off” that gene and its resultant disease.
Though its proof remains in animal studies, CRISPR’s ease and speed of use have made it a popular tool in laboratories worldwide in the short few years since the discovery of a method for using the Cas9 enzyme as a delivery system for gene editing in 2012. In one study on acute myeloid leukemia, a team led by Saar I. Gill, MD, PhD, an assistant professor of Hematology-Oncology, employed CRISPR to remove certain pieces of DNA from blood cells that CAR T cell therapy “hunters” target and home in on to destroy. The idea is that there will be less collateral damage to normal, healthy cells, resulting in lower toxicity in patients. The technique was proven effective in the lab on rodent and primate models, as well as in human cells.
Musunuru has been working in gene editing for more than 10 years. His work to permanently lower cholesterol in people genetically disposed for high levels of it has led to the founding of a start-up, Verve Therapeutics, in May. And while PCSK9 is the gene commonly targeted for cholesterol, even by drugs currently on the market, Musunuru and his team are focusing on new targets like ANGPTL3. About one in 300 people have a variant that naturally turns off this gene. “People with that variant have lower cholesterol levels, along with lower risk of type 2 diabetes and atherosclerosis,” Musunuru said.
That’s part of the game for researchers using CRISPR: finding a naturally occurring mutation and then chasing it to see whether it is actually linked to a condition and whether it’s viable to use.
For example, Musunuru’s lab has sought out gene variants in heart muscle cells that might be especially well equipped to stand up to a chemotherapy drug called doxorubicin, which has been tied to heart failure in some patients. By taking a pool of heart muscle cells in the lab and delivering CRISPR to them so that each has a different gene turned off, Musunuru’s work identified a number of potential gene editing targets that could factor into clinical trials within the space of a few years.
Musunuru is already using CRISPR to help patients by creating “genetic avatars”: stem cells engineered as a match to their own unique genetic sequences. Comparing these cells to typical stem cells can show whether the patient’s genetic variant is actually the cause of their disease. This knowledge can aid patients and their families in genetic counseling.
“In terms of research, CRISPR has already been transformative,” Musunuru said. “In terms of therapeutic applications, we are still in very early days.”
While Musunuru is still working toward the larger vision of that one-shot CRISPR treatment for high cholesterol, there is no question CRISPR’s clinical days are upon us. Penn Medicine is home to the first clinical trial to use CRISPR to edit human cells outside of China, where early experiments with CRISPR testing in humans have prompted a raft of bioethics debate across the globe. The new study, for patients with multiple myeloma, sarcoma, and melanoma, builds upon Penn’s world renown for the development of the first approved CAR T cell therapy, as well as the institution’s support for innovation in applying new technologies. Edward Stadtmauer, MD’83, a professor of Hematology Oncology, is leading the study in which patients’ hunter T cells are modified via CRISPR and then infused back into the body.
To move into the clinic for testing in patients, Stadtmauer’s research team had to progress through a gauntlet of institutional and federal regulatory approval steps that spanned more than two years. Now that the trial is underway, the researchers hope to improve upon early successes using CAR T cells, which have proven highly effective at treating some of the most intractable blood cancers, using this method that can so precisely fine-tune the genetic makeup of immune cells and easily edit multiple genes at once.
Results from Stadtmauer’s CRISPR modified T cell trial will be presented at the American Society of Hematology meeting in December 2019.