Drug-Target Interaction Prediction with Graph Regularized Matrix Factorization.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Experimental drug-target interaction (DTI) determination is costly and slow.
• Computational methods using drug-target networks aim to predict novel interactions.
• Existing algorithms struggle with predicting interactions for new drugs or targets (orphan nodes).

Purpose of the Study:

• To develop advanced computational methods for accurate drug-target interaction prediction.
• To address the challenge of predicting interactions involving novel drugs and targets.
• To improve the prediction of missing or unknown drug-target interactions.

Main Methods:

• Proposed two matrix factorization methods incorporating graph regularization to learn low-dimensional manifolds.
• Developed a preprocessing step to enhance predictions for new drugs/targets by adding edges with intermediate likelihood scores.
• Utilized a drug-target network (bipartite graph) as input for the algorithms.

Main Results:

• The proposed methods outperformed three state-of-the-art methods in cross-validation experiments.
• Graph Regularized Matrix Factorization (GRMF) demonstrated reasonable accuracy in predicting left-out interactions for simulated new drugs and targets.
• The graph regularization approach effectively learned underlying data manifolds.

Conclusions:

• The novel matrix factorization methods with graph regularization offer improved accuracy for drug-target interaction prediction.
• These methods show promise in addressing the challenge of orphan nodes in drug-target networks.
• The approach enhances the prediction of interactions for new drugs and targets, aiding drug discovery efforts.