GCNMDA

GCNMDA predicts associations between human microbes and drugs using a graph convolutional network framework that integrates microbe similarity, drug similarity, and microbe–drug interaction data.

Key Features:

• Heterogeneous Network Construction: GCNMDA constructs a heterogeneous network integrating a microbe similarity network, a drug similarity network, and a microbe-drug interaction network.
• Graph Convolutional Network (GCN): GCNMDA applies a graph convolutional network to learn node representations from the heterogeneous network for association prediction.
• Conditional Random Field (CRF) Integration: Conditional Random Fields are incorporated within GCN hidden layers to encourage similar nodes to have similar representations.
• Attention Mechanism in CRF Layer: An attention mechanism inside the CRF layer aggregates neighborhood representations to refine feature extraction.
• Random Walk with Restart Scheme: GCNMDA employs a random walk with restart on drug and microbe similarity networks to capture informative features.

Scientific Applications:

• Drug discovery acceleration: GCNMDA can prioritize candidate microbe-drug associations to accelerate drug discovery efforts.
• Drug repurposing facilitation: Predicted associations can identify opportunities for drug repurposing.
• Microbiome-pharmacology insights: GCNMDA enhances understanding of microbial roles in pharmacology by revealing potential microbe-drug interactions.
• Benchmark performance: GCNMDA demonstrated superior performance compared to seven state-of-the-art methods across three datasets.
• Case studies: Case studies involving SARS-CoV-2 and the antimicrobial drugs Ciprofloxacin and Moxifloxacin validated predicted microbe-drug associations.

Methodology:

GCNMDA constructs a heterogeneous network from microbe similarity, drug similarity, and microbe-drug interaction data; applies random walk with restart on similarity networks to capture features; processes the network with a graph convolutional network; and integrates Conditional Random Fields within GCN hidden layers with an attention mechanism in the CRF layer to aggregate neighborhood representations for prediction.