by Ildikó Ungvári, Gábor Hullám, Péter Antal, Petra Sz. Kiszel, András Gézsi, Éva Hadadi, Viktor Virág, Gergely Hajós, András Millinghoffer, Adrienne Nagy, András Kiss, Ágnes F. Semsei, Gergely Temesi, Béla Melegh, Péter Kisfali, Márta Széll, András Bikov, Gabriella Gálffy, Lilla Tamási, András Falus, Csaba Szalai
Genetic studies indicate high number of potential factors related to asthma. Based on earlier linkage analyses we selected the 11q13 and 14q22 asthma susceptibility regions, for which we designed a partial genome screening study using 145 SNPs in 1201 individuals (436 asthmatic children and 765 controls). The results were evaluated with traditional frequentist methods and we applied a new statistical method, called Bayesian network based Bayesian multilevel analysis of relevance (BN-BMLA). This method uses Bayesian network representation to provide detailed characterization of the relevance of factors, such as joint significance, the type of dependency, and multi-target aspects. We estimated posteriors for these relations within the Bayesian statistical framework, in order to estimate the posteriors whether a variable is directly relevant or its association is only mediated.
With frequentist methods one SNP (rs3751464 in the FRMD6 gene) provided evidence for an association with asthma (OR?=?1.43(1.2–1.8); p?=?3×10-4). The possible role of the FRMD6 gene in asthma was also confirmed in an animal model and human asthmatics.
In the BN-BMLA analysis altogether 5 SNPs in 4 genes were found relevant in connection with asthma phenotype: PRPF19 on chromosome 11, and FRMD6, PTGER2 and PTGDR on chromosome 14. In a subsequent step a partial dataset containing rhinitis and further clinical parameters was used, which allowed the analysis of relevance of SNPs for asthma and multiple targets. These analyses suggested that SNPs in the AHNAK and MS4A2 genes were indirectly associated with asthma. This paper indicates that BN-BMLA explores the relevant factors more comprehensively than traditional statistical methods and extends the scope of strong relevance based methods to include partial relevance, global characterization of relevance and multi-target relevance.