Biomedical Interaction Prediction with Adaptive Line Graph Contrastive Learning

Biomedical interaction prediction is essential for the exploration of relationships between biomedical entities. Predicted interactions can help researchers with drug discovery, disease treatment, and more. In recent years, graph neural networks have taken advantage their natural structure to achieve great progress in prediction. However, most them use node embedding instead directly using edge embedding, resulting information loss. Moreover, they predict links based on similarity correlation assumptions, which poor generalization. addition, do not consider difference topological negative positive sample links, limits performance. Therefore, this paper, we propose an adaptive line contrastive (ALGC) method convert into two kinds nodes. By adjusting number intra-class edges inter-class edges, augmented generated and, finally, views balanced by learning. Through experiments four public datasets, it proved that ALGC model outperforms state-of-the-art methods.