DHCLHAM: microbe-drug interaction prediction based on dual-hypergraph contrastive learning framework with...

Area of Science:

• Computational biology
• Pharmacology
• Microbiome research

Background:

• Drugs can disrupt gut microbiota, leading to adverse health outcomes.
• Current experimental methods struggle to elucidate microbe-drug interaction mechanisms.
• Existing computational approaches inadequately represent complex microbe-drug interactions.

Purpose of the Study:

• To develop an advanced computational framework for predicting microbe-drug interactions.
• To establish theoretical foundations for personalized and precision medicine in microbiome-drug therapies.
• To overcome limitations of graph-based methods in representing heterogeneous microbe-drug relationships.

Main Methods:

• Introduced a hierarchical attention-driven dual-hypergraph contrastive learning framework (DHCLHAM).
• Integrated nonlinear features, functional similarity (chemical attributes, microbial genomes), and Gaussian kernel similarity.
• Employed a dual network structure (KNN hypergraph, KO hypergraph) with hierarchical attention for information aggregation.
• Utilized contrastive learning to enhance heterogeneous hypergraph representation and multi-head attention for prediction score derivation.

Main Results:

• The DHCLHAM model significantly outperformed current optimal models on benchmark datasets.
• Achieved high performance metrics, including AUC and AUPR (e.g., 98.61% AUC and 98.33% AUPR on the aBiofilm dataset).

Conclusions:

• Developed a validated computational framework integrating AI and network pharmacology for microbe-drug interaction analysis.
• The findings provide a theoretical basis for optimizing clinical treatments and developing precise medication strategies targeting the gut flora.