Brian Hie wasn’t destined to be a scientist. During college, the computer science major was passionate about poetry, diving deep into the work of 17th-century poet John Donne. And when it came time to go to graduate school, he agonized over the decision between two diverging paths: English at Harvard and computer science at MIT. “There’s very few times where you have a set date yet to commit to a program, and you’re committing to different career paths,” said Hie.
Computer science prevailed — but in a way, Hie’s work still hinges on analyzing language. As a science fellow at alma mater Stanford and a visiting researcher at Meta AI, he uses machine learning techniques derived from natural language processing to understand protein evolution. “In natural language, your alphabet is maybe 50,000 words in a dictionary or something,” said Hie. “And then in biology, your alphabet will be 20 amino acids or it’ll be four DNA nucleic acids.”
Using so-called protein language models that were trained on massive repositories of protein sequences, Hie’s work has shown that algorithms can predict the evolution of proteins over geologic eons. Now, “we can show that these language models understand past evolution,” said Hie, “but what about using them to design new proteins?” His latest work focuses on using models to move evolution forward artificially, designing antibodies so they bind better to their targets.
Today, Hie’s confident he made the right call to stick with science — but he still likes to know that he can catch up on the latest Renaissance poetry when he wants to. “Thankfully,” he said, “humanities scholarship moves a lot slower than machine learning.”
— Katie Palmer
