Ralph investigated how protein language models, similar to ChatGPT but trained on protein sequences, can identify mutations that improve protein function, with the long-term goal of engineering a DNA polymerase to incorporate artificial nucleotides for therapeutic aptamer development. He developed a machine-learning framework to compare different protein language model representations across two well-characterized benchmark proteins. His findings showed that focusing on functionally important regions improved predictions of beneficial mutations, offering a promising approach to guiding the engineering of proteins with new functions.
I chose to conduct research abroad to expand my skills beyond the experimental protein engineering and computational chemistry methods I learned at UC Berkeley. Osaka University’s Institute for Protein Research stood out for uniting experimental, structural, and computational approaches in a single setting. Working in Professor Kenji Mizuguchi’s Laboratory of Computational Biology allowed me to learn machine-learning approaches to protein engineering while collaborating directly with international scientists. What surprised me most was how genuinely global the team was. I had always understood that science crosses borders, but seeing researchers from around the world bring distinct perspectives to shared problems gave that idea real weight. Experiencing that environment firsthand showed me how deeply modern research depends on international collaboration.
Ralph Vincent Mariano