3D Structure-based Generative Small Molecule Drug Design: Are We There Yet?

This is a benchmarking study for current 3D structure-based generative models, focusing on the chemical plausibility of generated structures. Novel metrics were proposed to facilitate the development of this area.

Dependencies

Conda environment:

A conda environment file is provided to install the conda environment.

conda env create -f environment.yml
conda activate cheminfo

Package requirement

The versions of key packages used in this study are shown as following:

Package	Version
Python	3.11.9
rdkit	2023.9.6
useful-rdkit-utils	0.56

Data availability

All the datasets used in metrics are provided, the ring system frequency of ZINC20 and ZINC22 drug-like molecules are under data folder. The BM scaffold files extracted from ZINC20 and ZINC22 drug-like molecules could be downloaded from Google Drive.

Structure of the repo

Molecules generated by each algorithm

The molecules generated by each algorithm are saved in folders named after the respective algorithms. All metrics and known medicinal chemistry filters are applied to each set of molecules in the Ring_system.ipynb file located within each folder.

Figure generation

All figures presented in the manuscript were generated using the Picture_drawing.ipynb file.

Contact

Feel free to create an issue or email Bo Yang ([email protected]) if you have any questions!