A Unified Protein Embedding Model With Local and Global Structural Sensitivity
Jerry Xu, 17, of Lexington, developed an AI model to rapidly compare the structures of proteins and amino acids for his Regeneron Science Talent Search computational biology and bioinformatics project.
Molecular biologists can learn a lot from a molecule鈥檚 structure, but identifying all the parts of that structure 鈥� as well as their purposes 鈥� is a massive computational task. Current methods either focus on overall structural comparisons, potentially missing key details in protein shape, or are too slow to handle massive databases of molecules.
In his project, Jerry found that structural information could be compressed into numerical strings for more efficient comparison without the loss of important features. His model could help biologists to learn about the function of currently unknown parts of proteins.
The son of Li Xu and Hao Ding, Jerry attends Lexington High School, where he is captain of both the math team and the science bowl club. Jerry co-founded a summer science lecture series for elementary through high school students, where he gives lessons on biology and math and performs live experiments. He also helps students with programming homework as a teaching assistant for KTBYTE Computer Science Academy.
Beyond the Project
For the past two years, Jerry has led a team of student coders in developing and maintaining a free chatbot to help town residents access information about their municipal budget.
FUN FACTS: Jerry is a big fan of the tabletop game 鈥淢agic: The Gathering.鈥� He enjoys the challenge of shaping his strategy with limited funds and options.
