DeepGCN-RT

News

We have released the training code and all trained model weights. If you find them helpful, please cite our article!

Kang, Q.; Fang, P.; Zhang, S.; Qiu, H.; Lan, Z. Deep Graph Convolutional Network for Small-Molecule Retention Time Prediction. Journal of Chromatography A. 2023, 464439, ISSN 0021-9673.

Article url: https://authors.elsevier.com/a/1hy7U4-ggV0Qt (https://doi.org/10.1016/j.chroma.2023.464439.)

Model performance

The model performances were evaluated through metrics including mean absolute error (MAE), median absolute error (MedAE), mean absolute percentage error (MAPE), mean square error (MSE), and R2.

Model performance

	Depth	MAE		MedAE		MAPE		R2		MSE
		Mean	Std	Mean	Std	Mean	Std	Mean	Std	Mean	Std
DeepGCN-RT	3	27.97	0.20	14.01	0.07	0.035	0.000	0.892	0.002	3303	55
DeepGCN-RT	5	27.00	0.19	12.91	0.18	0.034	0.000	0.892	0.001	3288	33
DeepGCN-RT	8	26.61	0.09	12.44	0.05	0.034	0.000	0.892	0.001	3286	31
DeepGCN-RT	16	26.55	0.17	12.38	0.12	0.033	0.000	0.892	0.001	3299	45

Note

This repository contians GNN models for rentention time prediction, including DeepGCN-RT, and plain GCN model. The models.py dataset.py and train.py contain source code. The model weights are included in the model_path folder. The transfer_learning.py contains the transfer learning code(10 fold cross validation). The full results of transfer learning for all models are contained in folder named result.

Build conda environment

The environment dependencies for Linux system are contained in the file named environment.yaml. Use conda update to build the environment.

Run the code

To run the training code, the following command could be used:

python train.py \
--model_name "DeepGCN-RT" \
--dataset "SMRT"
--num_layers 16 \
--hid_dim 200 \
--epochs 200 \
--lr 0.001 \
--batch_size 64\
--early_stop 30 \
--seed 1 

Alternatively, the train and transfer learning processes could also be started by the shell scripts using for loop. To run the training process on SMRT data set, using the following command:

sh scripts/train.sh

To run the transfer learning on nine transfer learning data sets, use:

sh scripts/transfer_learning.sh

Model inference using the DeepGCN-RT model.

To run the inference code, the following command could be used:

python inference.py \
--SMILES "demo_SMILES"\
--model_path "model path"