GRGMF

GRGMF predicts potential links in biomedical bipartite networks by applying graph-regularized generalized matrix factorization to learn latent node representations informed by neighborhood and external affinity data.

Key Features:

• Generalized Matrix Factorization Model: Employs a generalized matrix factorization model that captures latent patterns and learns node representations while adaptively incorporating neighborhood information.
• Graph Regularization: Introduces two graph regularization terms that leverage affinity information derived from external databases to refine latent node representations.
• New-node Handling: Integrates external affinity data to mitigate limitations for nodes that lack known link information.
• Performance Evaluation: Demonstrates competitive link-prediction performance in extensive experiments on six real biomedical bipartite datasets.
• Computational Compatibility: Implemented for CPU and CUDA execution, with a preference for CUDA for enhanced computational efficiency.

Scientific Applications:

• Biomedical Link Prediction: Predicts potential associations between entities such as genes and diseases or drugs and targets in bipartite networks.
• Drug Target Discovery: Identifies candidate drug–target associations and potential therapeutic pathways.
• Disease Research and Personalized Medicine: Supports identification of novel connections relevant to diagnosis, treatment, and personalized medicine.

Methodology:

Applies generalized matrix factorization to learn latent node representations from direct link information and neighborhood context and incorporates two graph regularization terms using external affinity data to refine predictions and address nodes without prior links.