Factors affecting the effective number of tests in genetic association studies: a comparative study of three PCA-based methods

Abstract

The number of tested marker becomes numerous in genetic association studies (GAS) and one major challenge is to derive the multiple testing threshold. Some approaches calculating an effective number (M eff ) of tests in GAS were developed and have been shown to be promising. As yet, there have been no comparisons of their robustness to influencing factors. We evaluated the performance of three principal component analysis (PCA)-based M eff estimation formulas (M effÀC in Cheverud (2001), M effÀL in Li and Ji (2005), and M effÀG in Galwey (2009)). Four influencing factors including LD measurements, marker density, population samples and the total number of tested markers were considered. We validated them by the Bonferroni's method and the permutation test with 10 000 random shuffles based on three real data sets. For each factor, M effÀC yielded conservative threshold except with D¢ coefficient, and M effÀG would be too liberal compared with the permutation test. Our results indicated that M effÀL based on r 2 coefficient achieve close approximation of the permutation threshold. As for a large number of markers, we recommended to use M effÀL with r 2 coefficient according to fixed-length separation, as well as fixed-number separation, to obtain accurate estimate of the multiple testing threshold and to save more computational time.

A Comparison of Methods for Gene-Based Testing That Account for Linkage Disequilibrium

“…On the other hand, combining the Bonferroni method with the estimate of Li and Ji (2005) did perform adequately and, as mentioned above, may be of interest when the signal is concentrated in a single SNP. Our findings are in line with those by Wen and Lu (2011) who showed that the method by Li and Ji (2005) performs better than other effective number of tests adjustments.…”

“…The statistical properties (i.e., type I error rate and power) of these methods have been examined in previous research ( Lin, 2005 ; Conneely and Boehnke, 2007 ; Chapman and Whittaker, 2008 ; Johnson et al, 2010 ; Moskvina et al, 2011 ; Wen and Lu, 2011 ; Alves and Yu, 2014 ). However, there are still several points that have not been considered so far.…”

Microbiota of the seminal fluid from healthy and infertile men

“…We controlled family wise error rate using Wen and Lu’s approach (48). Wen and Lu use the number of principle components, resulting from principal components analysis (PCA), that explain most of the variation in the data to estimate the effective number of independent hypothesis tests to control for.…”

1000 Genomes-based imputation identifies novel and refined associations for the Wellcome Trust Case Control Consortium phase 1 Data

“…The fact that we only have genome-wide significant signals from four wellestablished regions suggests that our conservative threshold fulfilled its purpose. Future studies may gain additional power with more sophisticated methods to control type-I error [27][28][29][30] or with methods that handle multiple related phenotypes. 31,32 On the other hand, for four out of the seven traits, our 1000 genomes-based imputation detected nothing novel on top of the original WTCCC study, suggesting that the potential power of imputation is limited by the genetic architecture of the trait(s) of interest and the genomic coverage of the GWAS genotyping panel used.…”