Deep Learning for Protein Design

using deep neural networks to refine unreliable regions of protein models

Check out the full lightning talk here!

This project encompasses my research with my undergraduate mentor Nao Hiranuma in the Institute for Protein Design at UW. We worked on a deep residual network to predict unreliable regions in computationally-derived protein structures (DeepAccNet), and applied another deep residual network to refine these unreliable regions (ULRs).

The DeepAccNet to Estogram (estimated divergence from true structures) to unreliable region (ULR) pipeline.
The ULR refinement pipleine, based on the DeepAccNet architecture. By focusing exclusively on ULRs and including surrounding regions with reliable structures, we improved the