Publication

Abstract : Designing reaction predictors for synthesising novel molecules require knowledge of the processes and structural comparisons of the molecules and the metabolic pathway classes. A hybrid graph-based deep learning model consisting of Graph Convolutional Network (GCN), and a convolutional neural network is proposed in this study to predict the metabolic pathway classes of the molecule. Also, appropriate changes are made to the hybrid model architecture to adapt multi-class classification. Unlike previously used machine learning approaches, the proposed model extracts important shape features which are atom-bond-based specifications using RDKit library from SMILES representations of the chemical molecule. The extracted features are then used to perform multi-class classification of molecular compounds by classifying them into several KEGG pathway classes. It was also found that the shape features generated by the GCN architecture can be used to predict the values of the top features for a given molecule. Finally, we compared the models including logistic regression, K-nearest neighbor, random forest, and our proposed hybrid model. The results show that the model performed well with an f1 score of 92.28%.