Identification of human protein subcellular location with multiple networks

Abstract

Protein function is closely related to its location within the cell. Determination of protein subcellular location is helpful to uncover its functions. However, traditional biological experiments to determine the subcellular location are of high cost and low efficiency, which cannot meet today’s needs. In recent years, lots of computational models have been set up to identify protein subcellular locations. Most models used features derived from protein sequences. Recently, features extracted from protein-protein interaction (PPI) network become popular to study various protein-related problems. A novel model with features derived from multiple PPI networks was proposed to predict protein subcellular location. Protein features were obtained by a new designed network embedding algorithm, Mnode2vec, which was a generalized version of the classic Node2vec algorithm. Two classic classification algorithms: support vector machine and random forest, were employed to build the model. Such model provided good performance and was superior to the model with features extracted by Node2vec. Also, this model outperformed some classic models. Furthermore, Mnode2vec can produce powerful features when the path length was small. The proposed model can be a powerful tool to determine protein subcellular location and Mnode2vec can efficiently extract informative features from multiple networks.