Modular Flows: Differential Molecular Generation

This is the implementation of Modular Flows: Differential Molecular Generatin NeurIPS 2022 paper, For more information visit the website

Prerequisites

• torchdiffeq : https://github.com/rtqichen/torchdiffeq.
• pysmiles
• rdkit-pypi==2021.3.5
• pytorch == 1.10.0
• torch-scatter
• torch-sparse
• torch-cluster
• torch-spline-conv
• torch-geometric
• astropy

The environment file Modflow_env.yml has been attached, to replicate the necessary environment to run the method.

Datasets

Datasets are placed in "data/" folder.

Usage

Different scripts are provided for different datasets. To see all options, use the -h flag.

Training:

QM9/ZINC:

python train_modflow_EGNN.py/train_modflow_EGNN_3D.py --nsamples #number_of_samples_use_to_train --data #dataset(QM9/ZINC) --batch_size #batch_size --niters #iterations

For, Tree representation we consider 30 common ring substructures. First, to create ring vocabulry and finding the common subsstructures, which will create files for ring vocabulary and common ring substructures.

python ring_index.py --data QM9/ZINC --nsamples #number_of_samples --nrings #number_of_rings_to_be_used
python train_modflow_EGNN_2D_JT.py/train_modflow_EGNN_3D_JT.py --nsamples #number_of_samples_use_to_train --data #dataset --batch_size #batch_size --nrings #number_of_rings --niters #iterations 

Evaluation/generation:

QM9/ZINC:

python testing.py/testing_3D.py --esamples #number_of_samples_to_generate --data #dataset  --model_name #name/location_of_model_in_Models_folder

For, Tree representation we consider 30 common ring substructures.

python testing_JT.py/testing_JT_3D.py --esamples #number_of_samples_to_generate --data #dataset --nrings #number_of_rings  --model_name #name/location_of_model_in_Models_folder

For tasks on toy-datasets,

python train_modflow_toy.py --nsamples #number_of_samples --data #dataset ("4x4_chess","16x16_chess","stripes") --batch_size #batch_size --niters #iterations

python testing_toy.py --esamples #number_of_samples --data #dataset ("4x4_chess","16x16_chess","stripes") --model_name #name/location_of_model_in_Models_folder

For Property Optimization,

python prop_optimize_qed.py --nsamples #number_of_samples --data #dataset --model_name #name/location_of_model_in_Models_folder