search

HUBioDataLab / DrugGEN

Official implementation of DrugGEN
GNU General Public License v3.0
48 stars 15 forks source link
deep-learning denovo-design drug-discovery drug-target-interactions generative-adversarial-network generative-models graph-neural-networks machine-learning pytorch transformer

readme

DrugGEN: Target Specific De Novo Design of Drug Candidate Molecules with Graph Transformer-based Generative Adversarial Networks

Updated Pre-print!

Please see our most up-to-date document (pre-print) from 26.07.2024 here: arXiv link

   

Abstract

Discovering novel drug candidate molecules is one of the most fundamental and critical steps in drug development. Generative deep learning models, which create synthetic data given a probability distribution, offer a high potential for designing de novo molecules. However, for them to be useful in real-life drug development pipelines, these models should be able to design drug-like and target-centric molecules. In this study, we propose an end-to-end generative system, DrugGEN, for the de novo design of drug candidate molecules that interact with intended target proteins. The proposed method represents molecules as graphs and processes them via a generative adversarial network comprising graph transformer layers. The system is trained using a large dataset of drug-like compounds and target-specific bioactive molecules to design effective inhibitory molecules against the AKT1 protein, which has critical importance for developing treatments against various types of cancer. We conducted further analysis with molecular docking and dynamics to assess the target-centric generation performance of the model, and attention score visualisation for interpretability. Results indicate that our de novo molecules have a high potential for interacting with the AKT1 protein at the level of its native ligands. Using the open-access DrugGEN codebase, it is possible to easily train models for other druggable proteins, given a dataset of experimentally known bioactive molecules.

Our up-to-date pre-print is shared here

   

Fig. 1. The schematic representation of the architecture of the DrugGEN model with powerful graph transformer encoder modules in both generator and discriminator networks. The generator module transforms the given input into a new molecular representation. The discriminator compares the generated de novo molecules to the known inhibitors of the given target protein, scoring them for their assignment to the classes of “real” and “fake” molecules (abbreviations; MLP: multi-layered perceptron, Norm: normalisation, Concat: concatenation, MatMul: matrix multiplication, ElementMul: element-wise multiplication, Mol. adj: molecule adjacency tensor, Mol. Anno: molecule annotation matrix, Upd: updated).

   

Transformer Module

Given a random molecule z, the generator G (below) creates annotation and adjacency matrices of a supposed molecule. G processes the input by passing it through a multi-layer perceptron (MLP). The input is then fed to the graph transformer encoder module Vaswani et al., (2017), which has a depth of 1 encoder layers with 8 multi-head attention heads for each. In the graph transformer setting, Q, K and V are the variables representing the annotation matrix of the molecule. After the final products are created in the attention mechanism, both the annotation and adjacency matrices are forwarded to layer normalization and then summed with the initial matrices to create a residual connection. These matrices are fed to separate feedforward layers, and finally, given to the discriminator network D together with real molecules.

   

Model Variations

   

Files & Folders

We provide the implementation of the DrugGEN, along with scripts from PyTorch Geometric framework to generate and run. The repository is organised as follows:

data contains:

experiments contains:

Python scripts:

   

Datasets

Two different data types (i.e., compound, and bioactivity) were retrieved from various data sources to train our deep generative models.

More details on the construction of datasets can be found in our paper referenced above.

   

Getting Started

DrugGEN has been implemented and tested on Ubuntu 20.04 with python >= 3.9. It supports both GPU and CPU inference.

Clone the repo:

git clone https://github.com/HUBioDataLab/DrugGEN.git

   

Training

Setting up environment

You can set up the environment using either conda or pip.

Here is with conda:

# set up the environment (installs the requirements):

conda env create -f DrugGEN/dependencies.yml

# activate the environment:

conda activate druggen

Here is with pip using virtual environment:

python -m venv DrugGEN/.venv
./Druggen/.venv/bin/activate
pip install -r DrugGEN/requirements.txt

Training the model

# Download input files:

cd DrugGEN/data

bash dataset_download.sh

cd

# Default DrugGEN model can be trained with the one-liner:

python DrugGEN/train.py --submodel="DrugGEN" --raw_file="DrugGEN/data/chembl_train.smi" --dataset_file="chembl45_train.pt" --drug_raw_file="DrugGEN/data/akt_train.smi" --drug_dataset_file="drugs_train.pt" --max_atom=45

** Explanations of arguments can be found below:

Model arguments:
  --submodel           Choose the submodel for training
  --act                Activation function for the model
  --max_atom           Maximum atom number for molecules must be specified
  --lambda_gp          Gradient penalty lambda multiplier
  --dim                Dimension of the Transformer model
  --depth              Depth of the Transformer model
  --heads              Number of heads for the MultiHeadAttention module
  --mlp_ratio          MLP ratio for the Transformers
  --dropout            Dropout rate for the encoder
Training arguments:
  --batch_size         Batch size for the training
  --epoch              Epoch number for Training
  --g_lr               Learning rate for G
  --d_lr               Learning rate for D
  --beta1              Beta1 for Adam optimizer
  --beta2              Beta2 for Adam optimizer
  --resume             Whether to resume training
  --resume_epoch       Resume training from this epoch
  --resume_iter        Resume training from this step
  --resume_directory   Load pretrained weights from this directory
  --set_seed           Whether to set seed
  --seed               Seed for reproducibility
  --use_wandb          Whether to use wandb for logging
  --online             Use wandb online
  --exp_name           Name for the experiment
Dataset arguments:      
  --features           Additional node features (Boolean) (Please check new_dataloader.py Line 102)

   

Molecule Generation Using Trained DrugGEN Models in the Inference Mode


python DrugGEN/inference.py --submodel="{Chosen model name}" --inference_model="DrugGEN/experiments/models/{Chosen model name}"

   

Deep Learning based Bioactivity Prediction


cd DEEPScreen2.1/chembl_31

python 8_Prediction.py AKT AKT

   

Results (De Novo Generated Molecules of DrugGEN Models)

The system is trained to design effective inhibitory molecules against the AKT1 protein, which has critical importance for developing treatments against various types of cancer. SMILES notations of the de novo generated molecules from DrugGEN models, along with their deep learning-based bioactivity predictions (DeepScreen), docking and MD analyses, and filtering outcomes, can be accessed under the paper_results folder. The structural representations of the final selected molecules are depicted in the figure below.

structures Fig. 2. Promising de novo molecules to effectively target AKT1 protein (generated by DrugGEN model), selected via expert curation from the dataset of molecules with sufficiently low binding free energies (< -8 kcal/mol) in the molecular docking experiment.

   

Updates

   

Citation

@misc{nl2023target,
    doi = {10.48550/ARXIV.2302.07868},
    title={Target Specific De Novo Design of Drug Candidate Molecules with Graph Transformer-based Generative Adversarial Networks},
    author={Atabey Ünlü and Elif Çevrim and Ahmet Sarıgün and Hayriye Çelikbilek and Heval Ataş Güvenilir and Altay Koyaş and Deniz Cansen Kahraman and Abdurrahman Olğaç and Ahmet Rifaioğlu and Tunca Doğan},
    year={2023},
    eprint={2302.07868},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

Ünlü, A., Çevrim, E., Sarıgün, A., Yiğit, M.G., Çelikbilek, H., Bayram, O., Güvenilir, H.A., Koyaş, A., Kahraman, D.C., Olğaç, A., Rifaioğlu, A., Banoğlu, E., Doğan, T. (2023). Target Specific De Novo Design of Drug Candidate Molecules with Graph Transformer-based Generative Adversarial Networks. arXiv preprint arXiv:2302.07868.

   

References/Resources

In each file, we indicate whether a function or script is imported from another source. Here are some excellent sources from which we benefit from:

Our initial project repository was this one.

   

License

Copyright (C) 2024 HUBioDataLab

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.