Interpretable machine learning algorithms reveal gut microbiome features associated with atopic dermatitis.

Area of Science:

• Microbiome research
• Computational biology
• Dermatology

Background:

• The gut-skin axis is implicated in atopic dermatitis development and symptoms.
• Quantitative screening of gut flora is essential for understanding this connection.

Purpose of the Study:

• To construct an interpretable machine learning framework for quantitatively screening key gut flora.
• To identify significant gut microbial features associated with atopic dermatitis.

Main Methods:

• Analysis of 16S rRNA datasets using five machine learning models (Random Forest, LightGBM, XGBoost, SVM, Logistic Regression).
• Application of centered log-ratio transformation and SHAP (SHapley Additive exPlanations) values for feature interpretability.
• Evaluation of model performance on validation partitions.

Main Results:

• Random Forest demonstrated superior performance among tree-based models.
• Bifidobacterium was identified as the strongest predictive factor for atopic dermatitis across all models.
• SHAP analysis revealed quantitative differences in gut microbiota composition.

Conclusions:

• Machine learning models integrated with SHAP values offer a method for quantitatively screening key gut flora in atopic dermatitis.
• This approach provides clinicians with an intuitive understanding of 16S rRNA sequencing data.
• The findings support the advancement of precision medicine for atopic dermatitis care and recovery.