MTMR: Molecule-to-Molecule Translation using Metric Learning and Reinforcement Learning

We propose MTMR, a molecular translation model based on metric learning and reinforcement learning, to achieve property improvement and high structural similarity performance at once.

MTMR can translate an existing drug into a novel drug candidate to empower desirable chemical properties without large changes of molecular structures.

MTMR requires molecule data represented by the format of simplified molecular-input line-entry system (SMILES) strings.

For more detail, please refer to Choi, Jonghwan, et al. "Collaboration of Metric Learning and Reinforcement Learning Enables Molecule-to-Molecule Translation for Drug Discovery" (under review)

• Latest update: 29 April 2022

---

SYSTEM REQUIERMENTS:

• MTMR requires system memory larger than 8GB.

• (if GPU is available) MTMR requires GPU memory larger than 8GB.

---

Installation:

• We recommend to install via Anaconda (https://www.anaconda.com/)

• After installing Anaconda, please create a conda environment with the following commands:

git clone https://github.com/mathcom/MTMR.git
cd MTMR
conda env create -f environment.yml

---

Data:

• Before running tutorials, an user should decompress the compressed files: DATA/{name}.tar.gz

• The following commands are for decompression:

cd DATA
tar -xzvf drd2.tar.gz
tar -xzvf qed.tar.gz
tar -xzvf logp04.tar.gz
tar -xzvf logp06.tar.gz
cd ..

• Due to the large size of the sorafenib dataset, please contact me if you need the dataset.

---

Tutorials:

• We provide several jupyter-notebooks for tutorials.

  • 1_pretraining.ipynb
  • 2_finetuning.ipynb
  • 3_latent_space_analysis.ipynb
  • 4_translation.ipynb
  • 5_evaluation.ipynb

• These tutorial files are available for reproducibility purposes.

• An user can open them using the following commands:

conda activate MTMR
jupyter notebook

~ run tutorial ~

conda deactivate

---

Contact:

• Email: [email protected]

---