ATOMIC: a graph attention network for atopic dermatitis prediction using human gut microbiome.

Area of Science:

• Computational biology
• Microbiome research
• Dermatology

Background:

• Atopic dermatitis (AD) is a chronic inflammatory skin disease with unclear etiology.
• Gut microbiota dysbiosis is implicated in AD pathogenesis, driving interest in microbiome-targeted therapies.
• Current computational models for disease prediction often lack interpretability and fail to capture complex microbial interactions.

Purpose of the Study:

• To develop an interpretable machine learning model for predicting atopic dermatitis (AD) using gut microbiome data.
• To incorporate microbial genomic information and co-expression networks for enhanced predictive accuracy.
• To identify key microbial taxa associated with AD for biomarker discovery and personalized interventions.

Main Methods:

• Developed ATOMIC, an interpretable graph attention network-based model.
• Integrated microbial co-expression networks with genomic information as node features.
• Trained and validated the model on 99 gut microbiome samples from adult AD patients and healthy controls.

Main Results:

• ATOMIC achieved high predictive performance, with an AUROC of 0.810 and AUPRC of 0.927.
• The model identified specific microbes associated with AD prediction.
• The interpretable attention mechanism highlighted key microbial taxa contributing to AD classification.

Conclusions:

• ATOMIC offers an interpretable approach to predict AD using gut microbiome data.
• The model facilitates the discovery of microbial biomarkers for AD.
• Findings support the development of personalized, microbiome-based interventions for atopic dermatitis.