Practical overview
When it fits
Use DrugEx when a research workflow begins with one or more desired scaffold fragments and needs model-based generation or multi-objective exploration of candidate molecules.
Who it is for
- Researchers exploring computational drug design from specified molecular scaffolds.
- Drug-design teams comparing multiple objectives in polypharmacology workflows.
Documented uses
- Prepare datasets for model pre-training and fine-tuning.
- Train a predictor that supplies rewards to a molecular-design agent.
- Train graph-transformer or SMILES-based molecular-generation models.
- Generate molecules with a trained model and analyze training performance.
Requirements & limits
The repository describes the software as free for academic users and directs users seeking commercial use to contact the authors. The repository lists an MIT license; review its license terms for the intended use.
- The documented dependency set includes Python 3.7 or later, PyTorch 1.7 and RDKit, alongside scientific Python packages.
- The repository documents separate graph-based and SMILES-based training paths.
- The README describes predicted rewards and generated molecules; it does not present these outputs as experimental validation.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare and train a DrugEx molecule-generation workflow
Follow the documented sequence for preparing data, training a reward predictor and training graph- or SMILES-based models.
Before you start
- Python 3.7 or later
- NumPy 1.19 or later, scikit-learn 0.23 or later, pandas 1.2.2 or later, PyTorch 1.7, Matplotlib 2.0 or later, and RDKit 2020.03 or later
- A GPU/CUDA setup appropriate to the device if using CUDA, as noted in the README
Steps
Run dataset.py to prepare data for pre-training and fine-tuning.
Run environ.py to train the predictor used to provide rewards.
Run train_graph.py to pre-train and train the graph-transformer model under supervised and reinforcement-learning frameworks, or use train_smiles.py for the SMILES-based models.
Expected outcome
A prepared dataset, trained reward predictor and trained graph- or SMILES-based model for the documented workflow.
Guide limitations
- The README’s subsequent molecule-generation stage uses designer.py with a well-trained model; this guide does not specify its arguments or input formats.
- CUDA toolkit selection depends on the GPU device.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- XuhanLiu/DrugEx · Official homepage title · retained in its source language