Matthew Liu L&S Math & Physical Sciences
Deep structure-based models for protein evolution
Understanding how and why proteins evolve is vital to grasping the fundamental biological processes governing everything from disease treatment to evolutionary biology. Classical sequence-based models of protein evolution make unrealistic assumptions ignoring behavior like coevolution of amino acids, which limits their explanatory power. Capturing such complex interactions requires a deep understanding of protein structure and how sites interact. To do this, my project introduces the first structure-based model of its kind for protein evolution, adapting models from protein design to contribute to a novel literature on deep models of protein evolution. I plan to design and implement a graph-attention model to predict evolutionary transitions by combining 1D sequence information with 3D structure information like bonding angles and site-site interactions. I then hope to investigate its effectiveness in downstream applications like variant effect prediction and ancestral sequence reconstruction.
Message To Sponsor
Thank you so much for sponsoring my project this summer! It's a huge privilege to get paid to do research full time, and I grew a lot this summer as a researcher because I had this opportunity. This has solidified my plans to pursue a PhD after I graduate, and it's been one of the best summers of my life.