Fig. 2.

Schematic illustration of compact feature learning. The Node2Vec algorithm is firstly used to calculate the topology information in complex networks. GIP kernel similarity and drug structure information are then extracted by a stacked automatic encoder, and the heterogeneous information is integrated to obtain a low-dimensional representation of the feature vector of each node. The resulting low-dimensional descriptor integrates the attribute characteristics, interaction information, relationship attributes and network topology of each protein or target node in the complex network