Using Machine Learning to Predict Cancer Treatment Response
Ramkumar Hariharan, associate teaching professor and director of the College of Engineering Programs in Seattle, discusses how he is developing machine learning tools to predict responses to cancer immunotherapy and democratize access to these treatments.
Ramkumar Hariharan, Northeastern faculty and program director at the Seattle campus, has embarked on a dual mission: democratize cancer treatments and enhance their efficacy by using machine learning. We sat down with Hariharan to learn about his work in geroscience, cancer immunotherapy, and the inspiration motivating his twofold aim.
Tell us about the inventions you’re creating right now and why they’re important.
At Northeastern’s Institute for Experiential AI, we’re focused on two projects. The first directly ties into research on aging and involves making a platform where users can upload their own data. The second focuses on predicting responses to cancer immunotherapy.
One of the themes in aging research looks at age-related diseases. Falling under geroscience, it involves connecting the link between aging and disease, while taking into account that aging is the single biggest risk factor for the onset of these age-related diseases. If we can slow down or halt aging, we could stave off the start of these diseases including cancer, diabetes, and heart disease. The molecular damage that happens with aging could be slowed down.
In current cancer immunotherapy treatments, there’s a chance there will be no recurrence for five to 10 years for solid tissue cancers (like breast cancer). These treatments work, but not all the time, meaning two or three out of every 10 patients might benefit.
What if we could predict the patients before they start immunotherapy? Can we find the biological response? If we can do that, we’re helping cancer immunotherapy reach far more patients. Machine learning (ML) is all about finding patterns in data. We can find data from existing patients and this data contains different patient’s age, gender, pre-existing conditions, stage of cancer, how bad it is, or if it has spread. We call these “differential features” in ML. The stage, age, weight, and other conditions are all ML features that will help predict whether a person will respond to cancer immunotherapy.
How will your innovation make a positive impact on the world for people from all walks of life?
It will be a nonproprietary tool anyone can use who has access to the internet. We are doing this work with an engineer at the Institute. He’s been building out the engineering pipeline, and I’m supplying the science in collaboration with other folks. I’m doing this to democratize our technology so it’s equitable, because that’s the right way to go about this. More people will find it useful, in turn we’ll get feedback, and we’ll keep improving the tool. This is academic science at its best – that’s the dream and hope.
What inspired you to bridge the gap between lab and world?
We had a cancer training center in my hometown in India. While working on my PhD research, I discovered that my lab was located right next to a pain and palliative care unit. I was brought into close contact with what suffering means and the toll it takes on people tending to a cancer patient. I also saw that a lot of the things we do in lab do not translate easily to patients. Translating any of the findings from the lab to the clinic was this huge, giant gap. I also thought that machine learning could help. If you have enough data, instead of reinventing the wheel we can get clues from the data. Some of the patients are so complex they are in 50 dimensions, so the question becomes “How can we see patterns?” That is why we need ML, and it’s why I turned to bridging the gap between clinical applications and data research.
On your journey to commercialization, what lessons have you learned?
Technology commercialization is the hardest. Working in the lab and on the computer gives you a lot of satisfaction. You think you have solved a problem, but then you scale and that’s when rubber hits the road. If you’re trying to put a healthcare product into the hands of consumers, that’s a totally different ball game. If you invent something, it’s less than 50% done. It’s a giant leap. One must always ask: “Will this make sense for the clients you’re trying to reach?”
Also, I never stop learning. I used to think I’d stop learning in my 30s, but I’ve been proven wrong every single day.
What are the benefits of being at Northeastern?
Northeastern is the place where I have found the least bureaucracy. Every single person in my department has sent me letters of personal acknowledgement and encouragement. It tells me that these things matter to the university. You’re also giving back to the university. The encouragement has been from almost everyone and that really makes me jump out of bed at 6 am.
I also do a lot of mentoring. My philosophy is to be a mentor and not a tormenter. That is incredibly important for data people. Motivate them first and be there for them when they need trouble shooting and guidance, but do so without killing their creativity.
Do you have any advice for students who dream of becoming inventors?
Do your best and pursue your passion. It’s age-old advice, but it has served me well.
New algorithms keep coming. Be wary of them, but also use them as tools to further your own interest.
Many people think it is human versus machine. It does not have to be this way; it can be human and machine.
Keep an open mind and be open to new tech advances. Tech is changing so fast. It is an era of AI, but also an era of so many other things. Use new tech and if it does not exist to solve your problem, invent new tech to solve your problem.
Source: Center for Research Innovation