Estimation of p-values with Reduced Set of Genomic Association Data

Permutation test has been in common use to get a statistical significance of an association measure when genomic association study is conducted. Its process involves repeated randomizing of the relationships between phenotype and genotype parts of the data as well as the calculation of the testing measure to form a null distribution point from each of the repetition. To compensate for the multiple comparison effect, which can result in overly optimistic p-values, the null distribution should be constructed by the maximum values among all the SNP combinations from each iteration. With increasing number of SNPs, especially when studying higher order gene-gene interactions, the amount of computation may soon become formidable. Possibility of dealing with reduced number of SNPs while producing the comparable result has been investigated in this paper. We found that, down to about 5% of the data considered, the estimated null distribution, verified by p-values were comparable to the one with original set of data.