Multimodal protective and susceptibility clusters in paediatric atopic dermatitis: A machine learning-based, data-driven observational study

Atopic dermatitis (AD) is a chronic inflammatory skin condition that often emerges in early childhood. Prevalence and allergy sensitisation patterns of AD differ between African populations, such as the AmaXhosa population in South Africa, and other groups, hinting at unique underlying mechanisms. Children from rural and urban communities of the AmaXhosa population, although sharing a common genetic ancestry, exhibit varying environmental exposures, making them an ideal subject for studying environmental and immune factors contributing to AD.

In a recent study, researchers employed machine learning (ML) techniques to examine a multimodal dataset containing environmental, cytokine, antibody, and transcriptomic data from healthy and AD-affected AmaXhosa children, aged 12–36 months, residing in either rural or urban areas. The objective was to uncover features linked to AD development by examining each data modality independently and then integrating them to discover multimodal signatures associated with the condition.

The environmental and antibody datasets revealed that the combined impact of environmental factors and elevated levels of allergen-specific and total IgE antibodies significantly contributed to AD prediction. Meanwhile, the transcriptomic dataset identified a group of 560 genes that differentiated children with and without AD, serving as the basis for further analysis. The integrated analysis then unveiled three distinct multimodal clusters associated with AD status.

One cluster comprised features indicative of a healthy phenotype, including environmental elements prevalent in rural settings, which correlated with plasma cytokine levels and the expression of autophagy-related genes. The other two clusters were defined by correlations between allergen-specific and total IgE antibodies, along with cytokine MCP-4 and TARC, and by a transcriptomic feature signature that was linked to the AD endotype.

These findings offer valuable insights into the multifaceted factors contributing to AD and provide a framework for future investigations into complex multimodal datasets in biomedical research.