Predicting Drug-Target Interactions With Multi-Information Fusion.

Area of Science:

• Computational Biology
• Pharmacology
• Bioinformatics

Background:

• Accurate prediction of drug-target interactions is crucial for drug discovery and repurposing.
• Existing computational methods face limitations, including oversimplified models and lack of negative samples in datasets.

Purpose of the Study:

• To develop a novel computational framework, NormMulInf, to overcome limitations in predicting drug-target interactions.
• To leverage labeled and unlabeled interaction data using semi-supervised learning and collaborative filtering theory.

Main Methods:

• Developed NormMulInf, a semi-supervised learning framework integrating biological information to determine sample and label similarities.
• Employed robust principal component analysis (RPCA) solved via augmented Lagrange multipliers to integrate similarity information.
• Utilized four classes of drug-target interaction networks for experimental validation.

Main Results:

• NormMulInf accurately classifies and predicts drug-target interactions across diverse networks.
• The method successfully predicted known interactions found in public databases.
• Demonstrated capability in predicting new drug targets (e.g., atropine with adrenergic receptors) and new drug-target associations (e.g., olanzapine, propiomazine with 5HT2B).

Conclusions:

• NormMulInf offers a robust approach to enhance drug-target interaction prediction.
• The framework addresses limitations in current methods, including the scarcity of negative samples.
• NormMulInf shows potential for advancing multitarget drug and multidrug target studies.