In this study, a Graph Convolutional Network (GCN) model was trained to predict the masked node class (Carbon: C, Nitrogen: N or Oxygen: O) using molecular graphs generated from FreeSolv dataset. Each molecule was repre- sented as a graph, with atoms as nodes and bonds as edges. The model achieved an overall accuracy of 0.84 on the hold-out test set for node classification task, demonstrating the superior capability of graph-based deep learning (DL) models like GCN in node level prediction task using molecular graphs as input.
For more details, please refer to this doc.
This project is intended strictly for learning and educational purposes. The paper (dummy) and model architecture (GCN) have been implemented only to understand Graph Neural Networks. The code should not be used for production, deployment, or any critical decision-making.