Multi-view heterogeneous molecular network representation learning for protein-protein interaction prediction.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions and signaling pathways.
• Incomplete data on PPIs hinders understanding of molecular mechanisms underlying human diseases.
• There is a critical need for computational methods to predict PPIs from a systems perspective.

Purpose of the Study:

• To develop an efficient computational model, MTV-PPI, for predicting protein-protein interactions (PPIs).
• To leverage heterogeneous molecular networks for simultaneous learning of protein sequences and molecular interactions.
• To enhance the understanding of molecular roots of human disease through accurate PPI prediction.

Main Methods:

• MTV-PPI utilizes a heterogeneous molecular network, learning inter-view protein sequences (k-mer method) and intra-view interactions (LINE embedding).
• Protein representation is constructed by aggregating inter-view and intra-view features.
• Random forest classifier is employed for the final prediction of potential PPIs.

Main Results:

• MTV-PPI achieved high performance on a heterogeneous molecular network, with 86.55% accuracy, 82.49% sensitivity, and 89.79% precision.
• The model demonstrated strong predictive power with an Area Under the Curve (AUC) of 0.9301 and Area Under the Precision-Recall Curve (AUPR) of 0.9308.
• Comparative experiments confirmed MTV-PPI's effectiveness over various protein representations and classifiers for PPI prediction in complex networks.

Conclusions:

• MTV-PPI is presented as a promising computational tool for predicting protein-protein interactions.
• The model offers a novel approach for future research in interaction prediction using heterogeneous molecular networks.
• This work contributes to advancing the field of systems biology and understanding complex molecular interactions.