Exploring Microbe-Drug Association Prediction via Multi-Attribute Dual-Decoder Graph Autoencoder.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Predicting microbe-drug associations (MDA) is crucial for understanding pathogenesis and developing targeted therapies.
• Traditional experimental methods for MDA prediction are time-consuming and labor-intensive.
• Existing graph embedding methods often lack comprehensive feature integration, limiting predictive accuracy.

Purpose of the Study:

• To develop an advanced computational method for predicting microbe-drug associations (MDA).
• To improve the accuracy of MDA prediction by integrating diverse data sources and graph properties.
• To provide a reliable tool for pharmaceutical innovation and targeted therapeutic development.

Main Methods:

• Constructed a heterogeneous network incorporating microbe similarity, drug similarity, and known associations.
• Enriched node information by importing topological features and multi-attribute data.
• Employed a dual-decoder graph autoencoder with a graph masking strategy to learn both node embeddings and graph structure.
• Integrated low-dimensional features and utilized a random forest classifier for final MDA prediction.

Main Results:

• The proposed MDGAEMDA method significantly outperformed existing advanced methods on public datasets.
• Experimental results demonstrated the model's superior performance in predicting microbe-drug associations.
• A case study involving widely used drugs validated the reliability and effectiveness of the MDGAEMDA approach.

Conclusions:

• MDGAEMDA offers a powerful and accurate computational approach for predicting microbe-drug associations.
• The method's ability to integrate multi-attribute features and graph structures enhances predictive capabilities.
• This work contributes to accelerating pharmaceutical innovation and advancing targeted therapeutics through improved MDA prediction.