Prediction of drug-target interactions from multi-molecular network based on LINE network representation method

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Experimental drug-target interaction (DTI) validation is costly and time-consuming, limiting the known DTI landscape.
• Existing computational methods often overlook the complex interplay between drugs, proteins, and other molecules.

Purpose of the Study:

• To develop an accurate and efficient computational model for predicting potential drug-target interactions.
• To leverage heterogeneous molecular data for improved DTI prediction.

Main Methods:

• Constructed a heterogeneous multi-molecular information network integrating data on proteins, drugs, lncRNAs, diseases, and miRNAs.
• Employed the Large-scale Information Network Embedding (LINE) model to learn node behavior information.
• Utilized a Random Forest classifier for training and predicting drug-target interactions based on integrated attribute and behavior information.

Main Results:

• Achieved 85.83% prediction accuracy and 80.47% sensitivity with an AUC of 92.33% via five-fold cross-validation.
• Case studies demonstrated significant success, with 8, 7, and 6 top-ranked candidate targets experimentally verified for Caffeine, Clozapine, and Pioglitazone, respectively.

Conclusions:

• The developed network embedding-based model is a powerful tool for predicting potential DTIs.
• This approach facilitates the identification of novel drug-target associations and potential therapeutic applications.