BNNR

BNNR performs bounded nuclear norm regularization-based matrix completion to predict drug–disease associations for drug repositioning.

Key Features:

• Matrix Completion Framework: Treats drug repositioning as a matrix completion problem under a low-rank assumption for the drug–disease association matrix.
• Bounded Nuclear Norm Regularization: Uses bounded nuclear norm regularization to construct low-rank approximations and to balance approximation error with rank properties while handling noise in similarity data.
• Incorporation of Constraints: Enforces interval constraints on predicted matrix entries to keep predicted interaction scores within realistic bounds.
• Heterogeneous Network Integration: Operates on an adjacency matrix derived from a heterogeneous network integrating drug–drug, drug–disease, and disease–disease associations, enabling use of all available relations and handling cold-start scenarios.
• Performance and Precision: Demonstrates higher prediction accuracy and particularly improved precision compared with existing state-of-the-art methods.

Scientific Applications:

• Drug Repositioning: Predicts novel drug–disease associations to identify new therapeutic uses for existing drugs.
• Drug Discovery Prioritization: Ranks candidate drug–disease links with high precision to support resource allocation in pharmaceutical research.

Methodology:

Formulates prediction as low-rank matrix completion solved with bounded nuclear norm regularization on an adjacency matrix from a heterogeneous network (drug–drug, drug–disease, disease–disease), incorporates drug–drug and disease–disease similarity matrices to handle noise, and applies interval constraints on predicted entries.