Using the longest significance run to estimate region-specific p-values in genetic association mapping studies
1 Department of Mathematics, National Changhua University of Education, Changhua 500, Taiwan
2 Institute of Biomedical Sciences, Academia Sinica, Taipei 115, Taiwan
3 Bioinformatics Program, Taiwan International Graduate Program, Academia Sinica, Taipei 115, Taiwan
4 Institute of Statistical Science, Academia Sinica, Taipei 115, Taiwan
5 Graduate Institute of Clinical Medical Sciences, Chang-Gung University, Taoyuan 333, Taiwan
BMC Bioinformatics 2008, 9:246 doi:10.1186/1471-2105-9-246Published: 27 May 2008
Association testing is a powerful tool for identifying disease susceptibility genes underlying complex diseases. Technological advances have yielded a dramatic increase in the density of available genetic markers, necessitating an increase in the number of association tests required for the analysis of disease susceptibility genes. As such, multiple-tests corrections have become a critical issue. However the conventional statistical corrections on locus-specific multiple tests usually result in lower power as the number of markers increases. Alternatively, we propose here the application of the longest significant run (LSR) method to estimate a region-specific p-value to provide an index for the most likely candidate region.
An advantage of the LSR method relative to procedures based on genotypic data is that only p-value data are needed and hence can be applied extensively to different study designs. In this study the proposed LSR method was compared with commonly used methods such as Bonferroni's method and FDR controlling method. We found that while all methods provide good control over false positive rate, LSR has much better power and false discovery rate. In the authentic analysis on psoriasis and asthma disease data, the LSR method successfully identified important candidate regions and replicated the results of previous association studies.
The proposed LSR method provides an efficient exploratory tool for the analysis of sequences of dense genetic markers. Our results show that the LSR method has better power and lower false discovery rate comparing with the locus-specific multiple tests.