GLDM_PyPi

This repository contains the user documentation of gldm package. Implementation of GLDM is stored here.

Environment Setup

gldm depends on pytorch, pytorch-geometric and rdkit libraries. To install the dependencies, please create a conda environment with the config file GLDM.yml:

conda env create --file=environment.yml
conda activate gldm

gldm can then be installed via pip:

pip install gldm

Adapt pytorch-geometric Package

We need to customize the pre-batching and mini-batching behaviors of pytorch-geometric. Please check the details in adapted_pkgs/pyg/README.md.

Inference with Pretrained Models

Download Pretrained Models

Our models were developed in two stages. First, the autoencoder models were trained to learn the latent space of the molecular graphs. Then, the diffusion models were trained to manufacture the latent space. The pretrained autoencoder models can be downloaded from here, and the diffusion models can be downloaded from here.

Usage

A simple example to generate 10 molecules with dummy gene expression profiles and save the generated molecules as samples.png:

from gldm import sampleMol
import torch

if __name__ == "__main__":
    model_path = 'GLDM_models/GLDM_WAE_cond.pt'
    config_file = 'config/GLDM_WAE_cond.yml'
    # gene_expression_file = 'pseudo_gene_expr_dose.pt'
    dummy_gene_expr = torch.rand((10, 979))  # 978 gene expression profile and 1 dosage value
    sampleMol(model_path, config_file, gene_expression=dummy_gene_expr, num_samples=10, output_file='samples.pkl', save_img='samples.png')

A similar example can be found in Example_usage.ipynb.

The sampleMol function has the following parameters:

Parameter	Description
model_path	The path to the pretrained model file.
config_file	The path to the configuration file.
gene_expression	A PyTorch tensor or a Numpy array or a list representing the gene expression + dosage.
gene_expression_file	The path to the file containing the gene expression tensors. If gene_expression is provided, this parameter will be ignored.
num_samples	The number of molecules to be generated.
output_file	The pickle file where the generated molecules will be saved.
save_img	The path where the generated image will be saved.

Note:

• The default configuration files are provided in the config folder. Remember to change config['model']['first_stage_config']['ckpt_path'] to the path that you stored the pretrained autoencoder models.
• We provide conditional and unconditional models using VAE, AAE and WAE losses. Please ensure the autoencoder model and the diffusion model are consistent in terms of the loss function and the conditional setting.
• Ensure the gene expression data are provided if the conditional models are used.

Training GLDM from scratch

Coming soon ...

---

Citation

@article{10.1093/bib/bbae142,
    author = {Wang, Conghao and Ong, Hiok Hian and Chiba, Shunsuke and Rajapakse, Jagath C},
    title = "{GLDM: hit molecule generation with constrained graph latent diffusion model}",
    journal = {Briefings in Bioinformatics},
    volume = {25},
    number = {3},
    pages = {bbae142},
    year = {2024},
    month = {04},
    issn = {1477-4054},
    doi = {10.1093/bib/bbae142},
    url = {https://doi.org/10.1093/bib/bbae142},
    eprint = {https://academic.oup.com/bib/article-pdf/25/3/bbae142/57160111/bbae142.pdf},
}