Machine learning-based prediction models for atopic dermatitis diagnosis and evaluation.

Area of Science:

• Dermatology
• Bioinformatics
• Computational Biology

Background:

• Atopic dermatitis (AD) is a prevalent chronic inflammatory skin condition impacting patient quality of life.
• Accurate diagnostic and treatment evaluation methods are crucial for effective AD management.

Purpose of the Study:

• To develop and validate machine learning (ML) models for novel AD diagnosis and therapeutic effect evaluation.
• To identify key genes associated with AD using integrated datasets and network analysis.

Main Methods:

• Candidate AD genes were identified using Robust Rank Aggregation (RRA) and protein-protein interaction (PPI) networks from four microarray datasets.
• Machine learning models (LASSO, Logistic Regression, Random Forest) were trained and tested on independent AD datasets (GSE130588, GSE99802).
• Model performance was evaluated using Area Under the Curve (AUC) and correlation analyses with clinical scores (SCORAD) and immune cell infiltration.

Main Results:

• The LASSO (REC) and Logistic Regression (REC and AAG) models demonstrated high accuracy in classifying AD lesions and non-lesions (AUCs ranging from 0.7761 to 0.8783).
• The LASSO (REC) and LR (AAG) models showed significant positive correlation with SCORAD and predicted immune cell infiltration.
• Model performance was validated across different treatment groups (Dupilumab, Crisaborole, fezakinumab).

Conclusions:

• Machine learning models provide a robust and accurate approach for AD diagnosis.
• These ML models can effectively evaluate the therapeutic effects of AD treatments.
• The developed predictive models offer a novel tool for clinical application in AD management.