Energy Basd Latent 3D Graph Diffusion

Overview

A pipeline to compress 3D graphs into the latent space, where a (vectorial) diffusion model is implemented to capture the distribution.

We added a seperated CNN that gets 3d latents as inputs and manipultates the latents in the way that it can minimize the prediciton energy.

Data

please Download all necessary data as it is provided in the original Latent 3D graph diffusion repository. Please also follow the same package installation steps as well.

Latent 3D graph diffusion

Training Energy based Latent

please go to the folder below

AE_Geometry_and_Conditional_Latent_Diffusion

and run the get_central_energy.ipynb notebook for training the energy part.

Energy spatial Map

Here is the absolute binding energy based on the displacement of the ligand from ground truth location.

Output files

the output Pickle file for the enrgy map and our CNN loss is available in the ./data/energy_grid.pkl and ./data/loss_z.pkl files.

Acknowledgements

We wanted to thank Dr. Yang Shen and Yuning You for their invaluable help throughout this project.