Molecular Optimization Model with Patentability Constraint (MOMP)

Installation:

1. Install conda or miniconda.
2. Set up the environment by running the following commands from the main folder:
   i. conda env create -f environment.yml
   ii. conda activate MOMP
3. Download the SureChEMBL dataset from: SureChEMBL Dataset
   i. Place the downloaded dataset in the dataset/SureChEMBL folder.
   ii. Run: python handle_patents_dataset.py from the main folder.
   iii. Move the generated SureChEMBL.txt file to dataset/SureChEMBL.

Note: Alternatively, to save time on preprocessing the data, you can download the patentsFp.pkl file from the following link: patentsFp.pkl. Place the downloaded patentsFp.pkl file in the dataset/PL/ folder.

Training:

From the main folder, execute the following command:

python train.py 2>error_train.txt

train.py is the main training file containing most of the hyper-parameters and configuration settings. After training, the checkpoints will be saved in the checkpoints folder.

Main Settings:

• epochs: Number of epochs to train the end-to-end model (QED=12 / DRD2=18).
• property: Name of the property to optimize - QED / DRD2.
• SR_similarity: Minimal similarity required for success (QED=0.15 / DRD2=0.15).
• SR_patentability: Maximum distance from patents for success (QED=0.4 / DRD2=0.6).
• SR_property: Minimal property value for success (QED=0.7 / DRD2=0.7).
• use_fp: Whether the translator uses molecule fingerprints (True/False).
• use_C_fp: Whether the translator uses molecule C-fingerprints (True/False).
• no_pre_train_models: Disable METNs pre-training (True/False).
• load_checkpoint: Load existing checkpoint (True/False).

Inference:

From the main folder, execute the following command:

python test.py 2>error_test.txt

Ablation Experiments:

Results of all the ablation experiments can be found in the ablation_outputs folder.

Baselines:

Results of all the baseline experiments can be found in the baselines_outputs folder.

Note: The MOMP model can be avoided from training, as the test files for the MOMP model with QED and DRD2 can be found in baselines_outputs\QED and baselines_outputs\DRD2, respectively.