Graph2MDA: a multi-modal variational graph embedding model for predicting microbe-drug associations.

Area of Science:

• Microbiology
• Pharmacology
• Computational Biology

Background:

• Human-associated microbes influence drug efficacy and toxicity.
• Identifying microbe-drug associations is crucial for drug research and development.
• Increasing genomic and pharmacological data necessitates effective computational methods.

Purpose of the Study:

• To develop a novel computational method for predicting microbe-drug associations.
• To leverage multi-modal data for improved prediction accuracy.
• To facilitate drug discovery by identifying new microbe-drug interactions.

Main Methods:

• Proposed Graph2MDA, a method utilizing variational graph autoencoder (VGAE).
• Constructed multi-modal attributed graphs integrating microbe and drug features (molecular structures, genetic sequences, function annotations).
• Trained VGAE to learn latent representations, followed by a deep neural network classifier for prediction.

Main Results:

• Graph2MDA demonstrated superior performance over six state-of-the-art methods on three independent datasets.
• Learned drug representations showed clustering consistent with ATC classification.
• Case studies validated 75-95% of predicted associations against PubMed literature.

Conclusions:

• Graph2MDA is an effective and robust method for predicting microbe-drug associations.
• The model's learned representations offer insights into drug properties and relationships.
• This approach significantly advances the screening of microbe-drug interactions for drug development.