Predicting microbe-drug associations with structure-enhanced contrastive learning and self-paced negative sampling...

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Predicting microbe-drug associations (MDAs) is crucial for drug development and precision medicine.
• Experimental methods for MDA discovery are time-consuming and costly.
• Existing computational methods, including graph contrastive learning (GCL), have limitations in capturing complex biological graph structures and selecting effective negative samples for training.

Purpose of the Study:

• To develop a novel computational approach, SCSMDA, for accurate prediction of microbe-drug associations (MDAs).
• To enhance the representation learning of microbes and drugs within biological networks.
• To improve the efficiency and accuracy of MDA prediction models by optimizing negative sample selection.

Main Methods:

• Constructed similarity and meta-path-induced networks for microbes and drugs.
• Employed a structure-enhanced contrastive learning strategy to refine node embeddings.
• Utilized a self-paced negative sampling strategy for training a Multi-Layer Perceptron (MLP) classifier.
• Predicted potential microbe-drug associations using the trained classifier.

Main Results:

• The SCSMDA approach demonstrated significantly improved performance on MDA prediction tasks across three public datasets.
• The structure-enhanced contrastive learning effectively captured rich structural information in biological graphs.
• The self-paced negative sampling strategy enhanced the classifier's ability to distinguish true associations.
• Case studies validated SCSMDA's effectiveness in identifying novel microbe-drug associations.

Conclusions:

• SCSMDA offers a powerful and accurate computational tool for predicting microbe-drug associations.
• The integration of structure-enhanced contrastive learning and self-paced negative sampling represents a significant advancement in MDA prediction.
• This methodology holds promise for accelerating drug discovery and advancing precision medicine.