Many problems in biology, such as predicting gene expression and protein folding, share a crucial common feature: dynamics of interest are governed by a complex energy landscape. Constructing these landscapes ab initio often times requires inaccessible data, such as accurate measurements of reaction rate parameters. On the other hand, we are often presented with a huge number of samples drawn from the landscapes, either from molecular dynamics simulations or single-cell RNA sequencing. This raises the possibility of empirically learning these landscapes through specifically designed deep learning architectures, which is the goal of this project.

I am participating in SuperUROP to pursue my interest in applying deep learning to understand the natural world. My previous experiences in computational chemistry and deep learning research were instrumental in establishing this interest. I hope to provide new methods for understanding gene expression and protein folding.