Graph Convolutional Neural Network for IC50 Prediction Model Using Amyotrophic Lateral Sclerosis Targets

IC50 prediction for neurodegenerative disorders like amyotrophic lateral sclerosis is crucial in biomedical studies. Traditional machine learning models use molecular descriptors for IC50 prediction where most of the descriptors created by different tools are irrelevant and undefined. Hence, graph convolutional neural network, a deep learning algorithm is used in this paper for building more accurate IC50 prediction model based on the structural properties of drug molecules in graph format. The work is implemented with 32 protein targets of amyotrophic lateral sclerosis disorder. IC50 prediction is made by collecting canonical SMILES and their corresponding IC50 values of 2100 drugs from the ChEMBL databases. Featurization and conversion of SMILES to molecular graphs are done by the Deepchem library. The library is used for dataset creation and model building. The results show that the proposed GCNN model with their fine-tuned hyperparameters achieves a prediction rate of 73%.