TADF-DL

Github repository of Effect of molecular representation on deep learning performance for prediction of molecular electronic properties by Jun Hyeong Kim, Hyeonsu Kim, Woo Youn Kim.

Table of Contents

• Environmental
• Data
• Train
• Test
• AGGNN

Environmental

• python>=3.7
• PyTorch=1.7.1
• RDKit>=2020.09.3
• descriptastorus

Data

Preprocessing

Move to data/preprocessing/. Run extract.py to get aromatic ring dataset.

We used the PubChem database included up to CID 139598315.

Dataset

Our dataset is available in data/ directory.

Basic data structure is shown below

ID    SMILES    HOMO    LUMO    E(S1)   E(T1)

For example,

id1   c1ccccc1    -5.6    -1.6    2.6   2.4
id2   Cc1ccccc1    -5.5    -1.5    2.7   2.3
...

Data file should be in src/model_directory/data/. Then, run dataset_divide.py to split dataset into training set and validation set, test set.

Train

There are 4 models for predict TADF-related properties in src/. Move to src/model_directory/train/(e.g. src/GCN/train/)

Run jobsctript_train.x

Train script is shown below.

python -u ../script/train.py \
--num_workers $NUM_WORKERS \
--batch_size $BATCH_SIZE \
--num_epochs $EPOCH \
--lr $LR \
--lr_decay $LR_DECAY \
--save_every $SAVE_PER_EPOCH \
--hidden_dim $HIDDEN_DIM \
--N_GCN_layer $N_GCN_LAYER \
--N_predictor_layer $N_PREDICTOR_LAYER \
--N_properties $N_PROPERTIES \
--dropout $DROPOUT 1> ./results/log.txt

An explanation of the options can be found in src/model_directory/script/train.py.

or

Run python src/model_directory/script/train.py --help.

You can check train loss and validation loss in scr/model_directory/train/result/log.txt.

Test

Run jobsctript_test.x

Test script is shown below.

python -u ../script/test.py \
--batch_size $BATCH_SIZE \
--test_file $TEST_FILE \
--restart_file $RESTART_FILE \
--hidden_dim $HIDDEN_DIM \
--N_GCN_layer $N_GCN_LAYER \
--N_predictor_layer $N_PREDICTOR_LAYER \
--N_properties $N_PROPERTIES \
--dropout $DROPOUT

An explanation of the options can be found in src/model_directory/script/test.py.

or

Run python src/model_directory/script/test.py --help.

You can check test loss in scr/model_directory/train/test_results.txt.

AGGNN

AGGNN code is available in https://github.com/edvardlindelof/graph-neural-networks-for-drug-discovery