Predicting Missing Links in Gene Regulatory Networks Using Network Embeddings: A Qualitative Assessment of Selective Embedding Techniques

Due to the expensive nature and incomplete experimental data, it is always not feasible to experimentally identify exhaustively interrelationship among biological macromolecules. Link prediction is to computationally guess missing relations within a partially constructed network. Representation learning with graph embedding recently achieves great attention towards in-depth network structural analysis and graph entity prediction. In this work, we try to evaluate a few well-known graph embedding techniques for inferring links in gene regulatory networks. We consider random walk (RW) and graph neural network (GCN)-based embedding methods. It is worth mentioning that RW embedding techniques are not equipped with inherent link prediction capabilities. We try to make the infer missing network links. Experimental results show superior performance by GCN-based methods in comparison with RW-based embedding methods for missing link prediction.