Peter Frazier solves information collection problems.
He is working with researchers at Georgetown University to find a drug that can stop Ewing's sarcoma from metastasizing. "It's an operations research problem because you have this resource which is your time and money and the effort that you put toward doing these tests and you want to allocate that resource as efficiently as possible," says Frazier.
A common first step in drug discovery is to have robots quickly run simple tests on a huge number of chemical compounds selected at random. When a compound shows promise, researchers must choose which of its myriad relatives to test more extensively. "I'm trying to help them to decide which derivatives to test," says Frazier, "so they can allocate their effort wisely in order to maximize the possibility that they'll actually find the drug for this disease."
Taking into account variables like toxicity, bioavailability, and what's already known about each compound, Frazier calculates which derivatives would be best to test. "There's a way of assigning probabilities using Bayesian statistics to how likely each potential outcome is based on your prior belief—before you test a compound, you have some idea how good it is," he says. "That's how you are able to deal with the fact that you're valuing something that hasn't happened yet, because you can assign probability distributions."
In a sense, Frazier is automating what people do naturally. "It's just a way of quantifying what is the hunch and based on that hunch, what should I do," he says.
While subjective, Frazier says his method is useful in drug discovery where resources are limited. "If you don't make some kind of assumptions, you just won't be able to say anything at all and that's unacceptable," he says. "Bayesian statistics is a quantitative way of encapsulating your intuition about a problem. You're going through a haystack looking for the needle, so you want to be able to use the intuition that you have, even though it's not very reliable, to direct the search."
Frazier's methods, which also have applications in simulation optimization, perform markedly better than comparable methods, and he's working on a way to compare their success rate to that of people. "We think that it will improve things and that given a particular case you'll have a better chance of finding a drug that will be better than what you would have found otherwise, but I don't really know how to quantify that yet," he says. "If you can change it from a five percent success rate to a ten percent success rate, then you've doubled the number of drugs discovered."
Ever since he was a kid, says Frazier, he's wanted to be a professor. He chose Cornell for its collaborative culture. "It seems to be easier to do interdisciplinary work here than other placesfr and that's important to me because my work is statistics, optimization, biology, simulation, all kind of different stuff," he says.
If you see someone riding a mountain unicycle on campus, there's a good chance it's Frazier. "It's fun to be different. You ride down the street, and people will say 'Hello.' to you and say 'Oh that's cool! How do you do that?' and kids love it," he says. "Kids always smile and laugh and say 'Oh look, it's a guy on a unicycle!' and so I like that about it."
Miaolan Xie is a Ph.D. candidate in operations research and information engineering at Cornell University, where she is fortunate to be advised by Professor Katya Scheinberg. She obtained her Bachelor...
Read more about Handling messy real-world data
When Caroline He ’23 started at Cornell Engineering, she listed her area of academic interest as biomedical engineering. This is not surprising, as several of her grandparents and her father are...
Read more about A perfect fit: ORIE and Caroline He