Gating-Enhanced Hierarchical Structure Learning\nin Hyperbolic Space and Multi-scale\nNeighbor Topology Learning in Euclidean Space for Prediction of Microbe-Drug\nAssociations

Identifying drug-related microbes may help us explore\nhow the microbes\naffect the functions of drugs by promoting or inhibiting their effects.\nMost previous methods for the prediction of microbe-drug associations\nfocused on integrating the attributes and topologies of microbe and\ndrug nodes in Euclidean space. The heterogeneous network composed\nof microbes and drugs has a hierarchical structure, and the hyperbolic\nspace is helpful for reflecting the structure. However, the previous\nmethods did not fully exploit the structure. We propose a multi-space\nfeature learning enhanced microbe-drug association prediction method,\nMFLP, to fuse the hierarchical structure of microbe and drug nodes\nin hyperbolic space and the multiscale neighbor topologies in Euclidean\nspace. First, we project the nodes of the microbe-drug heterogeneous\nnetwork on the sphere in hyperbolic space and then construct a topology\nwhich implies hierarchical structure and forms a hierarchical attribute\nembedding. The node information from multiple types of neighbor nodes\nwith the new topological structure in the tangent plane space of a\nsphere is aggregated by the designed gating-enhanced hyperbolic graph\nneural network. Second, the gate at the node feature level is constructed\nto adaptively fuse the hierarchical features of microbe and drug nodes\nfrom two adjacent graph neural encoding layers. Third, multiple neighbor\ntopological embeddings for each microbe and drug node are formed by\nneighborhood random walks on the microbe-drug heterogeneous network,\nand they cover neighborhood topologies with multiple scales, respectively.\nFinally, as each scale of topological embedding contains its specific\nneighborhood topology, we establish an independent graph convolutional\nneural network for the topology and form the topological representations\nof microbe and drug nodes in Euclidean space. The comparison experiments\nbased on cross validation showed that MFLP outperformed several advanced\nprediction methods, and the ablation experiments verified the effectiveness\nof MFLP’s major innovations. The case studies on three drugs\nfurther demonstrated MFLP’s ability in being applied to discover\npotential candidate microbes for the given drugs.