Internal Coordinate Net (ICoN) for Sampling Conformational Ensembles of Highly Dynamic Proteins via Generative Deep Learning

1. ICoN is trained on ~10000 of fully atomistic and highly flexible conformations

2. It uses vector internal coordinate representation as input features- vBAT

3. It can train ~10K conformations in a few mins, and generate ~100K conformations in less than a min.

Dependancies

• python 3.8 >
  • pytorch - for Deep Learning
  • MDAnalysis - for trajectory I/O
  • pytraj - for analysis

Installation

• Clone the repository: git clone https://github.com/chang-group/ICoN.git
• Navigate to corresponding folder for further instruction

Instuctions for users

• If users want to use this package for their system, they can download the aB-crystallin57-69 folder containing three different folders (output, src, and visual). Please follow these steps:

1. Make a folder (i.e., ICON_Model) and copy the downloaded folder (aB-crystallin57-69) into it. (you can change the name of both folders accordingly)
2. Make a folder (i.e., TRAJ) within the ICON_Model folder to copy the trajectory .dcd and topology files .prmtop. Please go to the aB-crystallin57-69 folder for further details of each step.