In December of 2012, Su-In Lee was working at the University of Washington in Seattle. One day, she got the kind of call everyone fears: Her father, Cheol Lee, was diagnosed with gallbladder cancer, a rare and aggressive form of the disease. By the time he was diagnosed, his cancer had already spread to other organs.
“It was a surprise to our family,” Lee said on the newest episode of GeekWire’s Health Tech Podcast. “The next day I just went back to Korea and met his doctor and then found that it’s incurable, especially in that stage.”
Not only was his cancer incurable, there actually isn’t a single drug designed to treat it. So Cheol’s doctors gave him a drug designed for pancreatic cancer, the organ next to the gallbladder. They hoped it would give him a few more weeks with his family.
“While I was taking care of my father, I kept thinking that it would be really great if we can find a drug that works for his cancer,” Lee said. “After three months, he passed away.”
Lee was changed by her father’s passing. She came back to Seattle with a new goal: Find a way to help cancer patients get the best treatment possible, even if that only means giving them a little bit more time with their family.
In the past five years, Lee — a professor of computer science and genome science at the UW — has grown that goal into a precision medicine program called MERGE. MERGE uses machine learning along with a patient’s DNA and other health data to predict which treatments will work best to help them fight the disease.
It’s just one of the growing number of precision medicine projects that are leveraging an expertise in artificial intelligence and a growing understanding of the human body to fight diseases of all shapes and sizes, from deadly cancer to the mysterious Alzheimer’s disease.
Hear Lee’s full story and learn more about precision medicine work in the most recent episode of GeekWire’s Health Tech Podcast. Listen in the player below or search for “Health Tech” to subscribe in your favorite podcast app.
Lee was singularly equipped to meet the challenge she set out for herself. She’s an expert in computational biology, using advanced computer science and biological data to understand the human body.
“So I kept thinking that if we understand the genetic and the molecular profile of my father’s individual cancer, we can potentially find the drug that’s going to work the best for him,” Lee said. “Even if it’s stage four. For patients in that stage, it’s really extending several months of the lifetime — it means a lot, to the patient and to the family.”
Heather Mefford, a pediatric neurologist and the deputy scientific director of Seattle’s Brotman Baty Institute for Precision Medicine, said that the approach isn’t just about finding the treatment that will extend the patient’s lifetime. There are many reasons precision medicine could be valuable to a patient.
“To decrease their risk for side effects, for example. To maximize the effect that the therapy that you give them is going to have at either treating or preventing a particular disease,” she said. It’s all about “making healthcare as efficient as possible for the individual.”
The approach is also helping scientists better understand the causes of some diseases, like Alzheimer’s.
But Lee wanted to focus on cancer patients. She teamed up with two Seattle researchers and hematologists: Dr. Pamela Becker and Dr. Tony Blau, whose startup All4Cure was featured on a previous Health Tech episode.
Blau and Becker are hematologists, or blood cancer experts, so they focused on understanding acute myeloid leukemia, AML. It’s one of the most common kinds of leukemia.
To train the MERGE algorithm, Lee and her team extracted genetic profiles from 30 blood samples taken from Blau and Becker’s past patients. Thanks to new technology, we can create genetic profiles like this in just under a day.
But the amount of data it creates is immense. One person’s genetic profile can hold a terabyte of data. If you printed it into the average-sized paperback, that would be three million books, or about 20 percent of the Library of Congress.
That’s where machine learning comes in. Lee and her team used the genetic data from those patients — along with data on the drugs they took and how they responded — to train a machine learning algorithm. They also incorporated publicly available data on AML patients to expand the training set.
As it sorted through the data, the algorithm found patterns and started to make connections between certain genes and drugs. Interestingly enough, it made too many connections — Lee said teaching the algorithm how to tell the difference between important and unimportant connections was actually challenging.
“If this gene does something important in cancer, then that means that this association between that gene and in any drugs should be considered important,” Lee said. Genes that regulate tear ducts, for example, probably aren’t as important.
Today, MERGE is close to accurately predicting responses to some of the most common AML drugs. Lee and her team are continuing their work and even working on a more complex version of the system that can predict responses to multiple drugs at once.
Their work has been built on the data of real patients who were willing to share it anonymously in the name of science. Lee said her work, and the work of others, is dependent on the generosity of anonymous patients all over the world.
“I’ve been on both sides,” Lee said. “Patients really should consider giving their data. It’s like donating money — it can be better than that, actually. … I also got the data from my father’s sample. It wasn’t actually easy, you know — being a patient, I’m a little more emotional than being a scientist.”
She shared the data with the hospital where her father was treated. She wanted them to use it for research, in the hopes that one day it could lead to new treatments or cures for people like her father.
Editor’s note: Researcher Suman Jayadev‘s work is also included in this episode.