KronRLS-MKL

KronRLS-MKL predicts drug–target interactions by applying Kronecker Regularized Least Squares and multi-kernel learning to integrate heterogeneous biological data for link prediction on bipartite networks.

Key Features:

• Kronecker Regularized Least Squares (Kronecker RLS): Uses Kronecker RLS for link prediction on bipartite drug–target networks.
• Multi-Kernel Learning (MKL): Automatically selects and weights multiple kernels to optimize predictive performance.
• Kernel importance weighting: Assigns importance weights to each kernel to identify the most contributory biological information sources.
• Integration of heterogeneous data: Incorporates diverse biological data types to enhance prediction scope and accuracy.
• Scalability: Handles large-scale bipartite networks representing drugs and targets.
• Implementation: Provided as a Matlab implementation.
• Empirical evaluation: Evaluated across four datasets using twenty kernels and compared to 18 competing methods, achieving higher or comparable predictive performance.
• Exhaustive pairwise experiments: Uses exhaustive pairwise kernel experiments to reflect predictive quality of individual kernels.

Scientific Applications:

• Drug–target interaction prediction: Prioritizes likely interactions between drugs and protein targets via link prediction on bipartite networks.
• Drug lead prioritization: Supports identification and ranking of candidate drug–target pairs for follow-up studies.
• Data-source relevance analysis: Reveals biologically relevant information sources by weighting and selecting kernels according to predictive contribution.

Methodology:

Applies Kronecker Regularized Least Squares (Kronecker RLS) combined with multi-kernel learning (MKL) to automatically select and weight multiple kernels, evaluated via exhaustive pairwise experiments across four datasets using twenty kernels; implemented in Matlab.