What is it about?
Atopic dermatitis (AD) is less common in many developing countries and in rural environments, suggesting that environmental exposures interact with immune and molecular mechanisms to influence disease risk. The AmaXhosa population in South Africa provides a particularly interesting setting to study these relationships: children living in rural and urban environments share a common ethnogenetic background but experience very different environmental conditions. We used previously collected data from healthy AmaXhosa children and children with AD, combining information on living and health conditions with antibody and cytokine measurements and gene expression data. Using explainable machine learning, we first identified the features most strongly associated with AD within the questionnaire and antibody data. To enable the integration of the high-dimensional bulk transcriptomics data with the other data types, we also developed a software tool to identify a reduced set of significant and biologically meaningful transcripts. We then integrated all data types to uncover clusters of correlated environmental, immune and transcriptomic features associated with protection from or susceptibility to AD.
Featured Image
Photo by GuerrillaBuzz on Unsplash
Why is it important?
Large consortium projects increasingly collect multimodal data spanning environmental exposures, clinical characteristics and different molecular layers, but integrating these complex datasets remains a major challenge. Our study provides a practical framework for such integration, combining feature selection and explainable machine learning to identify meaningful relationships across data types. Importantly, the integrated analysis revealed clusters of correlated features across environmental, immune and gene expression data, linking both established and newly identified features associated with atopic dermatitis. This will be particularly interesting to investigate in future mechanistic studies. The approach may therefore be useful beyond atopic dermatitis for other complex diseases in which environmental, molecular and clinical factors interact.
Perspectives
From the outset, our goal was to make maximal use of an already established biomedical dataset by going beyond analysing the individual datasets separately and simply combining their results. We wanted to use advanced machine learning and multimodal data integration to uncover relationships that would otherwise remain hidden and gain a deeper understanding of the data. Achieving this required more than integrating different types of data. It brought together expertise spanning study design and sample collection, biomedical research, bioinformatics and machine learning, with people from these different fields working closely together to shape the analytical process and interpret and present the results. In this sense, we did not only integrate data, but also connected people across disciplines, with everyone contributing their respective expertise.
Katja Baerenfaller
Read the Original
This page is a summary of: Multimodal protective and susceptibility clusters in paediatric atopic dermatitis: A machine learning-based, data-driven observational study, PLoS Medicine, September 2026, PLOS,
DOI: 10.1371/journal.pmed.1004917.
You can read the full text:
Contributors
The following have contributed to this page
