/6.883ProteinDocking

Amanda Beck and Daniel Goodwin's CAPRI docking project

Primary LanguageJupyter NotebookApache License 2.0Apache-2.0

Predicting Protein Docking Sites

Amanda Beck and Daniel Goodwin Class project for 6.883, Autumn 2020

Overview

This code uses the ProteinGCN code as a base to work on the protein docking problem. Our hypothesis was tht the ProteinGCN network, trained on the Rosetta300k datset, would give a good embedding to learn the docking sites between two proteins. This hypothesis ended up being disproven, for our code, there was no performance difference between the embedded and non-embedded proteins.

To run the code

  • We convert the data into a loadable format using the notebook "Saving DB5 into loadable format"
  • We build the siamese/RelNet in "RelationNet on ProteinGCN to Predict Binding"
  • We explored using an autoencoder for a more global view of docking in "Final - AE Code"

Thanks

Big thanks to Iddo Drori, TA Zee Yan, and the class of 6.883 in Autumn 2020