DrugEx

The drugex package is organized into several functional subpackages that handle different stages of the drug discovery and machine learning workflow:

• drugex.data: Data management and handling.
• drugex.logs: Logging utilities.
• drugex.molecules: Molecular representation and manipulation.
• drugex.parallel: Parallel processing capabilities.
• drugex.training: Machine learning training workflows.
• drugex.utils: General utility functions.

DrugEx is a collection of deep learning models designed for the directed generation of molecules. To begin using the project, you should follow these primary steps:

1. Installation: Set up the environment and install the necessary dependencies.
2. Usage: Explore provided examples to understand how to run models for molecular generation.
3. API Documentation: Refer to the module-specific documentation for detailed technical specifications of the underlying software components.

The DrugEx tutorials are organized by model type. The recommended learning path is:

1. Recurrent Neural Network (RNN) (Sequence-RNN.ipynb): Recommended for beginners or those needing an introduction to generative AI for molecules.
2. Graph Transformer (Graph-Transformer.ipynb): An advanced tutorial that assumes knowledge of concepts covered in the RNN tutorial.

The drugex.molecules package is the core module for handling molecular data in DrugEx. It is organized into several functional submodules and subpackages:

• drugex.molecules.interfaces: Defines the abstract interfaces and base classes for molecular objects and suppliers.
• drugex.molecules.mol: Contains the primary implementations for molecular representations and data structures.
• drugex.molecules.suppliers: Provides modules for fetching or generating molecular data from various sources.
• drugex.molecules.converters (Subpackage): Contains logic for transforming molecular data between different formats or representations.

The drugex package provides several top-level modules for specific tasks:

• drugex.about: Package information and metadata.
• drugex.dataset: Tools for managing and interacting with datasets.
• drugex.download: Utilities for downloading external data or resources.
• drugex.generate: Functions for generating molecular data or structures.
• drugex.train: Modules for executing training pipelines.

Before running the CLI tutorials, you must download the necessary example data and pretrained models into the tutorial/CLI/examples directory. Run this command from the repository root:

python -m drugex.download -o tutorial/CLI/examples # ran from the repository root

To set up the environment for the DrugEx tutorials, install the required packages using pip from the provided requirements.txt file:

pip install -r requirements.txt

To follow the CLI tutorial, you must first install the drugex package and QSPRPred. Once installed, navigate to the /CLI directory and run the download command to populate the local environment with pretrained models, example data, and a local version of the Papyrus dataset.

This process creates an examples folder in your current directory and a data/.Papyrus folder containing the Papyrus dataset, which can be used for further data extraction via Papyrus-scripts.

drugex download -o examples
# or on Windows:
python -m drugex.download -o examples

Pretraining (-tm PT) follows the same workflow as fine-tuning, but the generator is not initialized with pretrained weights. This is used to train a model from the ground up.

# Preprocess
python -m drugex.dataset -b ${BASE_DIR} -i A2AR_LIGANDS.tsv -mc SMILES -o example_pt -mt graph

# Pretrain
python -m drugex.train -tm PT -b ${BASE_DIR} -i example_pt -mt graph -e 2 -bs 32 -gpu 0