Protective and Susceptibility Clusters of Environmental Factors, Gene Expression, Antibody Responses, and Cytokines in Pediatric Atopic Dermatitis: Insights from Multi-Modal Data Integration

BackgroundAtopic dermatitis (AD) is a chronic skin disease that typically occurs in early childhood. In this cross-sectional case-control study, our objective was to employ machine learning approaches to identify novel clusters of protective or susceptibility features associated with AD. Methods and FindingsWe utilised an integrated dataset comprising previously established environmental, cytokine, antibody, and gene expression data from AmaXhosa children, both healthy and with AD, living in either rural or urban settings of South Africa, aged 12-36 months. The applied machine learning methods included the GeneSelectR workflow to identify a subset of relevant genes, the calculation of SHAP values to explain the machine learning output, and the use of DIABLO to integrate the datasets for a comprehensive analysis. Key findings included the identification of a protective cluster of environmental features primarily found in the rural setting, which were correlated with plasma cytokine levels and with expression of autophagy-related genes. Additionally, we identified AD susceptibility clusters where levels of allergen-specific and total IgE antibodies correlated with the cytokines MCP-4 and TARC. Lastly, we identified an RNA-Seq feature signature specific to the disease endotype. ConclusionsThe application of various machine learning methods enabled the identification of significant factors associated with AD in a complex, multi-modular dataset, making the output explainable and potentially informing targeted interventions and improved diagnostic criteria.