Unified all-atom molecule generation with neural fields
This repository contains the model weights for FuncBind, a framework for target-conditioned 3D molecule generation using neural fields. As presented in the paper Unified all-atom molecule generation with neural fields, this unified model leverages score-based generative models and neural fields to represent molecules as continuous atomic densities, enabling it to be trained across diverse atomic systems and drug modalities.
Code: https://github.com/prescient-design/funcbind
This repository provides model weights for FuncBind and preprocessed datasets (train/test CrossDocked). The Macrocyclic Peptide Pair (MCP) dataset and its preprocessed splits are available at https://huggingface.co/datasets/Willete3/mcpp-dataset.
For detailed instructions on installation, data preparation, sampling, and training, please refer to the comprehensive GitHub repository.
Sample Usage
After setting up the environment and downloading the necessary checkpoints as outlined in the GitHub repository, you can sample macrocyclic peptides (MCPs) from the model using the following command:
python sample_fb.py --config-name sample_fb_mcpp