https://github.com/cmwoodley/diffdream
Science Score: 13.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
○CITATION.cff file
-
○codemeta.json file
-
○.zenodo.json file
-
✓DOI references
Found 4 DOI reference(s) in README -
○Academic publication links
-
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (5.5%) to scientific vocabulary
Repository
Basic Info
- Host: GitHub
- Owner: cmwoodley
- License: gpl-3.0
- Language: Jupyter Notebook
- Default Branch: master
- Size: 10.7 MB
Statistics
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
- Releases: 0
Metadata Files
README.md
"Diffream" - Pharmacophore guided generative diffusion model
Carried out as a side project during my PostDoc at the University of Liverpool to get to grips with the Pytorch library. Effectively an attempt at trying to replace the autoencoder used in LigDream with a 3D Unet and generate molecules using stable diffusion. More details are included in a presentation delivered internally at UoL.
Adapted a unet structure for medical image segmentation from 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation
Trained using
Implementation uses two methods of guidance for generation: - Pharmacophore nudging - Inputs noise to generate intermediate representation --> Pharmacophore representation predicted from intermediate rep. --> Calculate BCE loss with true pharmacophore --> adjust input with autograd --> SMILES predicted with captioning networks from Ligdream - Embedded pharmacophore conditioning - inputs noise and embedded representation of pharmacophore --> denoised for t timesteps mixing in with original sample --> generated representation captioned by SMILES
WIP
- Captioning networks are not good --> would like to replace with transformers
Owner
- Login: cmwoodley
- Kind: user
- Repositories: 1
- Profile: https://github.com/cmwoodley
GitHub Events
Total
Last Year
Dependencies
- diffusers ==0.12.1
- filelock ==3.9.0
- huggingface-hub ==0.12.1
- importlib-metadata ==6.0.0
- pyyaml ==6.0
- regex ==2022.10.31