Graph based link prediction for epilepsy drug discovery.

Area of Science:

• Computational Biology
• Neuroscience
• Pharmacology

Background:

• Epilepsy affects millions globally, presenting significant social and neurological challenges.
• Current understanding of epilepsy mechanisms is incomplete, hindering effective treatment development.
• Phytochemical-protein interactions, inspired by Ayurveda, offer a promising avenue for epilepsy therapy, especially against drug-resistant cases.

Purpose of the Study:

• To develop a computational framework for predicting phytochemical-protein interactions relevant to epilepsy treatment.
• To leverage graph-based approaches and Graph Neural Networks (GNNs) for this prediction task.
• To explore natural, Ayurveda-inspired alternatives for epilepsy management.

Main Methods:

• Modeled phytochemical-protein interactions as a bipartite graph.
• Employed Graph Convolutional Networks (GCN), Graph Attention Network (GAT), and GraphSAGE.
• Utilized one-hop enclosing subgraphs to refine predictive performance.

Main Results:

• The best-performing GNN model achieved high accuracy (0.9778), precision (0.9574), F1-score (0.9782), and ROC-AUC (0.9994).
• Graph-based computational methods proved effective and scalable for predicting interactions.
• The framework demonstrates potential for identifying novel epilepsy therapeutic candidates.

Conclusions:

• The proposed computational framework accurately predicts phytochemical-protein interactions for potential epilepsy treatments.
• Graph neural networks offer a powerful tool for drug discovery in neurological disorders.
• Ayurveda-inspired phytochemicals show promise for natural epilepsy therapies, validated by computational modeling.