Research Assistant
Statistical Physics Of Evolving Systems Lab At The University Of Washington
Seattle, Washington, United States
● Project 1: Collaborated with physicists and computer scientists to create HERMES. I curated protein structure datasets and wrote sections of the pre-processing pipeline. HERMES: Holographic Equivariant neuRal network model for Mutational Effect and Stability prediction. Gian Marco Visani, Michael N. Pun, William Galvin, Eric Daniel, Kevin Borisiak, Utheri Wagura, Armita Nourmohammad. bioRxiv 2024.07.09.602403; doi: https://doi.org/10.1101/2024.07.09.602403● Project 2: Co-built (with one colleague), trained, and tested a transformer-based deep learning model in Python (using PyTorch) to map protein structure to a representative low dimensional embedding.● Project 3: Helped develop convolutional neural network machine learning model in Python (using PyTorch) to learn protein side chain conformations from the protein backbone. Created an objective function to properly represent angle distance cost (180 as worst) and adapted the pre-processing pipeline to train on SO(3) equivariant representations of protein structure.● During these projects, I realized our Conda environments were consuming our allotted inodes on the UW computational cluster. I took initiative to improve our development process and created an Apptainer (similar to docker) build script to hide the environments in a single Apptainer image. I documented how to use the build script and image for my colleagues.● I used Git, Bash Scripting, Shell Utilities, and the Slurm Workload Manager on the Linux computational cluster.
