Please join us for this week’s CSB seminar this Wednesday 2/28, from 4-5pm in 32-124. Soojung Yang, a third-year CSB PhD student from Prof. Rafael Gomez-Bómbarelli’s group (DMSE), will be presenting her recent work on harnessing data augmentation for efficient exploration of protein conformational space. Pizza and drinks will be provided. All faculty, researchers, and students across departments and institutes are welcome!
Time & Place: Feb 28, 4-5pm in 32-124
Talk Title: Harnessing data augmentation for efficient exploration of protein conformational space
Abstract: Exploring rare events, such as protein folding, presents a computational challenge due to high energy barriers. If we can find good collective variables (CVs) that projects the protein system to a meaningful low-dimensional space, we can add bias along the CVs to flatten out the energy barrier so that the sampling is accelerated. However, without sufficient data from transition states, finding good CVs becomes problematic. How do we solve this chicken and egg problem? We propose to use a trick that’s popular in machine learning research – data augmentation with synthetic data. We show that the synthetic transition state samples generated from geodesic interpolation can contribute to CV search, ultimately enhancing sampling efficiency.
Speaker: Soojung Yang is a third-year PhD candidate in the CSB program, advised by Professor Rafael Gomez-Bómbarelli. Her research centers on sampling protein conformational ensembles through the integration of machine learning and molecular simulations.
