SPLNMTF

SPLNMTF applies self-paced non-negative matrix tri-factorization to integrate patient, gene, and drug data into a heterogeneous network for computational drug repositioning.

Key Features:

• Integration of Heterogeneous Biological Data: Integrates patient, gene, and drug data using non-negative matrix tri-factorization to capture cross-entity associations.
• Self-Paced Learning Mechanism: Implements a self-paced learning strategy with soft weighting that progressively incorporates samples from high-quality (easy) to low-quality (complex) to improve robustness and mitigate poor local optima.
• Tri-factorization to Enhance Learning Ability: Employs a tri-factorization framework to address learning ability deficiencies of traditional matrix factorization models, enhancing prediction of drug-disease associations.

Scientific Applications:

• Drug Repositioning Prediction: Predicts novel therapeutic uses for existing drugs and has been applied to identify candidate drugs for ovarian cancer and acute myeloid leukemia (AML).
• Comparative Performance Evaluation: Demonstrated performance that outperforms eight state-of-the-art models in the reported drug repositioning studies.

Methodology:

The approach constructs a heterogeneous network from patient, gene, and drug datasets, applies non-negative matrix tri-factorization to decompose the network, and incorporates a self-paced learning component that prioritizes high-quality samples via soft weighting.