Multilayer biological network alignment based on similarity computation via Graph Neural Networks

Abstract

Multilayer networks prove highly effective in scenarios where objects of different entities require a heterogeneity of types of interaction. Briefly, a multilayer network consists of nodes, edges, and layers, whose meaning depends on their entities. Its informative content can be studied to understand the biological interactions and mechanisms of a larger one, based on topological and homological similarities. This approach is referred to Network Alignment (NA), and it is generally used to transfer the knowledge from one biological species to another. From a more general point of view, NA allows mapping nodes between two or more networks of interest, by preserving topologically similar regions. In this paper, we presented MALGNN, a Multilayer-network ALigner based on Graph Neural Networks (GNN). It is a method for Pairwise Global NA (PGNA) of multilayer biological networks, which uses GNNs for processing node embeddings and computing the similarities between pairs of nodes. The proposed method performs the topological assessment via the unsupervised representational learning of the multilayer network graph models. MALGNN allows improving the alignment performance on multilayer networks, in terms of Node Correctness and Objective Score, compared to methods for static and dynamic/temporal networks. Our method proved to be a ready-to-use solution for performing the PGNA of multilayer networks based on topological assessment. Our experimentation demonstrated optimal performance in aligning multilayer networks, in terms of Node Correctness and Objective Score.