Fig. 2

The pipeline of ImpAESim algorithm. This framework mainly contains two parts, multi-network embedding to obtain a compact low-dimensional vector feature representation to describe the topological properties for each disease and disease similarity calculation based on distance measurement. First we integrate three disease-related information sources to construct three input networks (A), then we run RWR to learn global topological properties of the networks. The output of RWR is fed to the classic Auto-Encoder (B) to calculate the constraints and obtain low-dimensional vectors of hidden layer. Then the low-dimensional vectors and constraints are fed to the ImpAE (C) to obtain the low-dimensional representation of disease features after concatenating the hidden vectors. Finally the combined representations of diseases can be utilized to measure disease similarity by calculating a cosine distance (D)