A MAJOR genomic study has uncovered dozens of previously unknown asthma risk variants, using a combination of large-scale genetic analysis and deep learning to refine understanding of the disease.
Asthma is a common, highly heritable respiratory condition driven by complex interactions between genetic and environmental factors. While genome-wide association studies (GWAS) have identified many susceptibility loci, much of the disease’s polygenic architecture has remained unclear. The new research addressed this gap by integrating advanced statistical approaches with artificial intelligence to improve variant discovery and risk prediction.
Asthma Risk Variants Identified Through Multi-Method Approach
Researchers conducted the largest meta-analysis of asthma genetics to date in individuals of European ancestry, combining data from over 158,000 cases and more than 1.6 million controls.
To enhance discovery, they applied pleiotropy-informed approaches, including multi-trait analysis of GWAS and conditional false discovery rate, leveraging eosinophil count as a secondary trait. In parallel, a Transformer-based deep learning framework, InsightGWAS, was used to prioritise biologically relevant variants and refine downstream analyses.
This approach identified 69 previously unreported genome-wide significant loci associated with asthma risk. Additional candidate loci were uncovered through complementary statistical and deep learning analyses. Functional annotation suggested that these variants were linked to immune regulation, airway remodelling, and metabolic pathways, all key biological processes in asthma pathophysiology.
Deep Learning Improves Asthma Risk Prediction
The incorporation of deep learning provided an additional layer of variant prioritisation beyond conventional statistical methods. Polygenic risk score models derived from variants prioritised by the deep learning framework demonstrated improved predictive performance compared with those based on traditional GWAS approaches.
These findings suggest that integrating artificial intelligence into genomic pipelines may improve the accuracy of genetic risk stratification, with potential implications for earlier identification of individuals at higher risk of asthma.
The study was limited to individuals of European ancestry, which may restrict generalisability to other populations. As with all GWAS-based research, the findings also identify associations rather than direct causality.
Overall, the study provides a more comprehensive map of asthma genetics and highlights the growing role of deep learning in genomic medicine. These findings could support future efforts to develop personalised risk prediction tools and guide mechanistic research into asthma development.
Reference
Chen E et al. Deep learning and statistical methods identify novel asthma risk variants in Europeans. J Allergy Clin Immunol. 2026;DOI:10.1016/j.jaci.2026.03.016.
Featured image: Atlas on Adobe Stock
