DHRLS

DHRLS predicts gene-disease associations by combining Dual Hypergraph Regularized Least Squares with Centered Kernel Alignment-based Multiple Kernel Learning (CKA-MKL) to integrate multi-source biological kernels and capture higher-order relationships for link prediction in bipartite gene-disease networks.

Key Features:

• Multiple Kernel Learning: DHRLS integrates multiple kernels derived from diverse biological data sources for both genes and diseases.
• Hypergraph Regularization: DHRLS models higher-order relationships among genes and diseases using hypergraphs beyond pairwise interactions.
• Centered Kernel Alignment (CKA-MKL): CKA-MKL optimizes kernel selection and weighting in both gene and disease spaces.
• Alternating Least Squares Algorithm (ALSA): The DHRLS model parameters are solved using ALSA.
• Bipartite Link-Prediction Framework: Gene-disease association detection is framed as link prediction within a bipartite network.
• Performance Validation: The method demonstrated superior performance compared to existing prediction tools, validated with two cross-validation schemes and evaluated on six real-world networks.

Scientific Applications:

• Complex disease gene discovery: Predict and prioritize candidate genes associated with complex diseases.
• Integration of heterogeneous biological data: Combine multiple biological data types through kernel integration to improve association inference.
• Prioritization for experimental validation: Provide candidate gene-disease links to guide follow-up verification methods.

Methodology:

Construction of multiple kernels from various biological data sources for genes and diseases; application of CKA-MKL to determine and weight optimal kernels; utilization of hypergraphs to model higher-order relationships; solving the DHRLS model using ALSA to predict gene-disease associations.