RESUMO
Background: Asthma is a chronic inflammatory disease of the airways that is heterogeneous and multifactorial, making its accurate characterization a complex process. Therefore, identifying the genetic variations associated with asthma and discovering the molecular interactions between the omics that confer risk of developing this disease will help us to unravel the biological pathways involved in its pathogenesis. Objective: We sought to develop a predictive genetic panel for asthma using machine learning methods. Methods: We tested 3 variable selection methods: Boruta's algorithm, the top 200 genome-wide association study markers according to their respective P values, and an elastic net regression. Ten different algorithms were chosen for the classification tests. A predictive panel was built on the basis of joint scores between the classification algorithms. Results: Two variable selection methods, Boruta and genome-wide association studies, were statistically similar in terms of the average accuracies generated, whereas elastic net had the worst overall performance. The predictive genetic panel was completed with 155 single-nucleotide variants, with 91.18% accuracy, 92.75% sensitivity, and 89.55% specificity using the support vector machine algorithm. The markers used range from known single-nucleotide variants to those not previously described in the literature. Our study shows potential in creating genetic prediction panels with tailored penalties per marker, aiding in the identification of optimal machine learning methods for intricate results. Conclusions: This method is able to classify asthma and nonasthma effectively, proving its potential utility in clinical prediction and diagnosis.
RESUMO
BACKGROUND: Genome-wide association studies have yet to identify the majority of genetic variants involved in asthma. We hypothesized that expression quantitative trait locus (eQTL) mapping can identify novel asthma genes by enabling prioritization of putative functional variants for association testing. OBJECTIVE: We evaluated 6706 cis-acting expression-associated variants (eSNPs) identified through a genome-wide eQTL survey of CD4(+) lymphocytes for association with asthma. METHODS: eSNPs were tested for association with asthma in 359 asthmatic patients and 846 control subjects from the Childhood Asthma Management Program, with verification by using family-based testing. Significant associations were tested for replication in 579 parent-child trios with asthma from Costa Rica. Further functional validation was performed by using formaldehyde-assisted isolation of regulatory elements (FAIRE) quantitative PCR and chromatin immunoprecipitation PCR in lung-derived epithelial cell lines (Beas-2B and A549) and Jurkat cells, a leukemia cell line derived from T lymphocytes. RESULTS: Cis-acting eSNPs demonstrated associations with asthma in both cohorts. We confirmed the previously reported association of ORMDL3/GSDMB variants with asthma (combined P = 2.9 × 10(-8)). Reproducible associations were also observed for eSNPs in 3 additional genes: fatty acid desaturase 2 (FADS2; P = .002), N-acetyl-α-D-galactosaminidase (NAGA; P = .0002), and Factor XIII, A1 (F13A1; P = .0001). Subsequently, we demonstrated that FADS2 mRNA is increased in CD4(+) lymphocytes in asthmatic patients and that the associated eSNPs reside within DNA segments with histone modifications that denote open chromatin status and confer enhancer activity. CONCLUSIONS: Our results demonstrate the utility of eQTL mapping in the identification of novel asthma genes and provide evidence for the importance of FADS2, NAGA, and F13A1 in the pathogenesis of asthma.