High-speed westfall-young permutation procedure for genome-wide association studies

Abstract

Genome-wide association studies (GWASs) are widely used to investigate statistically significant associations between diseases and single nucleotide polymorphisms (SNPs) to identify causal factors of diseases. In GWAS, statistical significance of more than one million SNPs have been recently assessed, but in many case, no associations are found because of the application of conservative multiple testing corrections, such as Bonferroni correction. While more sensitive methods, such as Westfall-Young permutation procedure (WY), would relate more SNPs with diseases, its extremely long computational time has prohibited from the application of WY to GWAS. We introduce an algorithm to accelerate WY, named High-speed Westfall-Young permutation procedure (HWY). HWY utilizes three techniques to make WY computationally practical. First, P-value calculations for SNPs that cannot affect the adjusted significance level are pruned. Second, a lookup table of P-values is used to avoid frequent duplicate calculations. Finally, computations are parallelized using a GPGPU. HWY was 619 times faster than WY and more than 122 times faster than PLINK, a widely used GWAS software, and analyzed a dataset contained one million SNPs and one thousand individuals in approximately two hours. Re-analysis of existing GWAS datasets with HWY may uncover additional hidden SNP-trait associations.

AllerStat: Finding Statistically Significant Allergen-Specific Patterns in Protein Sequences by Machine Learning

“…The lower bound in (3) was proposed in FastWY method in (45,46), in which the authors studied itemset mining tasks to find statistically significant combinations of multiple genetic factors. In allerStat, we adapted the technique in the FastWY method for sequence mining setting.…”

“…In allerStat, we adapted the technique in the FastWY method for sequence mining setting. Therefore, the proof that the bound (3) satisfies the property in (2) can be simply done as described in (45,46).…”

“…The large number of candidate patterns raises not only a computational challenge but also a statistical one. Over the past decade, several methods have been proposed for evaluating the statistical significance of discovered patterns in several data mining tasks (33,(45)(46)(47)(48)(49). In this paper, we evaluate the statistical reliability of ASPs by adapting the techniques developed in these studies to our problem setup.…”

Alice and the Caterpillar: A More Descriptive Null Model for Assessing Data Mining Results

“…This direction is orthogonal to the study of the statistical validity of the results, which is our focus. Many works concentrate on the case of labeled transactional datasets (Terada et al, 2015(Terada et al, , 2013aPellegrina et al, 2019b;Hämäläinen, 2016;Pellegrina and Vandin, 2020;Papaxanthos et al, 2016;Minato et al, 2014;Llinares-López et al, 2015;Komiyama et al, 2017;Wu et al, 2016;Duivesteijn and Knobbe, 2011), where each transaction comes with a binary label. Most of these works use resampling-based approaches, as we do, but the very different nature of the studied tasks and data, as we study the unlabeled case, make them inapplicable to our problems.…”

A tutorial on statistically sound pattern discovery

“…In pattern discovery the most popular choices for evaluating dependency rules and other similar bipartition dependence patterns in the variable-based interpretation have been Fisher's exact test (e.g., (Hämäläinen, 2012;Terada et al, 2013bTerada et al, , 2015Llinares López et al, 2015;Jabbar et al, 2016)) and the χ 2 -test (e.g., (Morishita and Sese, 2000;Morishita and Nakaya, 2000;Nijssen and Kok, 2006;Hämäläinen, 2011;Jin et al, 2012;Terada et al, 2015)). Both of these tests have also been used for evaluating significance of improvement (see Section 4).…”