Variational Bayesian Multi-Kernel Adaptive Deep Fusion for Microbe-Related Drug Prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Discovering microbe-drug associations (MDAs) is crucial for drug discovery and understanding microbial mechanisms.
• Experimental methods for MDA identification are time-consuming and expensive, necessitating computational approaches.

Purpose of the Study:

• To develop an effective computational model for predicting novel microbe-drug associations (MDAs).
• To enhance the accuracy and efficiency of MDA prediction compared to existing methods.

Main Methods:

• Proposed a Variational Bayesian Multi-Kernel Adaptive Deep Fusion (VBMKADF) model.
• Integrated multiomics data to construct drug molecular graphs and microbe hypergraphs.
• Employed multilayer graph and hypergraph convolutions with an attention mechanism for similarity fusion, integrated into a Bayesian logistic matrix factorization framework.
• Utilized a variational Expectation-Maximization algorithm for adaptive inference and model training.

Main Results:

• VBMKADF demonstrated superior performance over state-of-the-art methods in predicting microbe-drug associations.
• Achieved higher AUPR, AUC, and F1 scores across balanced and imbalanced datasets.
• Case studies validated the model's effectiveness as a tool for MDA prediction.

Conclusions:

• The VBMKADF model provides a powerful and accurate computational tool for predicting microbe-drug associations.
• This approach accelerates drug discovery and deepens the understanding of microbial functions.
• VBMKADF offers a significant advancement over traditional experimental methods for MDA identification.