VB-MK-LMF: fusion of drugs, targets and interactions using variational Bayesian multiple kernel logistic matrix...

Area of Science:

• Computational biology
• Pharmacology
• Machine learning

Background:

• Computational fusion approaches improve drug-target interaction (DTI) prediction by leveraging multiple data sources.
• Optimizing DTIs requires careful consideration of observation weighting and side information focus.

Purpose of the Study:

• To introduce Variational Bayesian Multiple Kernel Logistic Matrix Factorization (VB-MK-LMF) for enhanced DTI prediction.
• To unify key advancements in DTI prediction, including multiple kernel learning and weighted observations.

Main Methods:

• Implemented multiple kernel learning, weighted observations, graph Laplacian regularization, and explicit probability modeling.
• Utilized variational Bayesian approximation for efficient computation on graphics processing units (GPUs).

Main Results:

• VB-MK-LMF significantly outperforms state-of-the-art methods on standard benchmarks.
• Analysis revealed benefits and limitations of linear kernel combinations and the impact of prior kernels on DTI tasks.
• Low-dimensional latent representations facilitate visual analytics, and probabilistic predictions enable estimation of pharmaceutically relevant quantities like drug promiscuity.

Conclusions:

• VB-MK-LMF demonstrates superior predictive performance across various settings in DTI prediction benchmarks.
• The study provides valuable estimates for drug promiscuity, druggability, and total drug-target interactions, advancing pharmaceutical research.