Computational Prediction of Drug-Disease Association Based on Graph-Regularized One Bit Matrix Completion

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug repositioning offers an efficient alternative to novel drug discovery for treating diseases.
• Computational methods can predict optimal indications for existing drugs using biological datasets.
• Drug-disease associations exhibit a low-rank structure, enabling matrix completion approaches.

Purpose of the Study:

• To propose a novel matrix completion framework for predicting drug-disease indications.
• To incorporate side information of drugs and diseases using a neighborhood graph.
• To develop an algorithm for binary data with graph regularization and score range constraints.

Main Methods:

• Graph regularized 1-bit matrix completion (GR1BMC) framework.
• Utilizes side information from drugs and diseases modeled as a neighborhood graph.
• Employs a parallel proximal algorithm to solve the constrained minimization problem.

Main Results:

• The GR1BMC framework was validated on two standard databases.
• Area Under the Curve (AUC) was evaluated using 10-fold cross-validation.
• A case study demonstrated successful prediction of top indications for novel drugs, verified with the CTD database.

Conclusions:

• GR1BMC is an effective computational method for drug-disease association prediction.
• The framework successfully integrates neighborhood information and handles binary data constraints.
• Validated predictions suggest utility in identifying new therapeutic indications for existing drugs.