moleculegen

Moleculegen-ML [1] is a Python package for de novo drug design based on generative language modeling. It comprises tools for molecular data processing, SMILES-based language modeling (recurrent networks, autoregressive transformers, convolutional networks) and transfer learning.

Documentation

For now, our wiki serves as a documentation (or rather a user guide) for the project. Our paper [1] is a survey of various machine learning methods for SMILES-based molecule generation.

If you find Moleculegen-ML useful in your research, please consider citing [1].

Installation

It is convenient to set up dependencies using environment management systems like conda or virtualenv. We use the latest stable version of Ubuntu to test our project.

Download, install, and set up Miniconda.

Create a new environment and install dependencies (see environment.yml and requirements.txt):

$ conda env create -f environment.yml
$ conda activate moleculegen

If you wish to run experiments on GPU (recommended), please install CUDA (we use version 10.1) and run

$ pip install -r requirements.txt

And finally, install the package:

$ pip install git+https://github.com/sanjaradylov/moleculegen-ml.git

Usage

See wiki for feature overview and documentation.

We provide the benchmarking script scripts/run.py. To observe the command line arguments print a help message:

$ python3 run.py --help

The paper uses standardized ChEMBL data from [4]. If you wish to experiment with your own dataset, consider also the preprocessing scripts in queries/ and scripts/.

References

1. Adilov, Sanjar (2021): Neural Language Modeling for Molecule Generation. ChemRxiv. Preprint. https://doi.org/10.26434/chemrxiv.14700831.v1
2. Segler et al. Generating Focused Molecule Libraries for Drug Discovery with Recurrent Neural Networks. https://arxiv.org/pdf/1701.01329.pdf
3. Gupta et al. Generative Recurrent Networks for De Novo Drug Design. DOI: 10. 1002/minf.201700111
4. Brown et al. GuacaMol: Benchmarking Models for de Novo Molecular Design. J. Chem. Inf. Model. 2019, 59, 1096−1108
5. D. Polykovskiy et al. Molecular sets (moses): a benchmarking platform for molecular generation models. 2020. Front Pharmacol 11:58.