Addressing Population‐Specific Multiple Testing Burdens in Genetic Association Studies

• Published in: Annals of human genetics Vol. 79; no. 2; pp. 136 - 147
• Main Authors: Sobota, Rafal S., Shriner, Daniel, Kodaman, Nuri, Goodloe, Robert, Zheng, Wei, Gao, Yu‐Tang, Edwards, Todd L., Amos, Christopher I., Williams, Scott M.
• Format: Journal Article
• Language: English
• Published: England Wiley Subscription Services, Inc 01.03.2015
• Subjects: Asian People - genetics; autocorrelation; Genetics, Population; Genome-Wide Association Study; Genome‐wide threshold; GWAS; HapMap; HapMap Project; Humans; linkage disequilibrium; Melanoma - genetics; Polymorphism, Single Nucleotide; White People - genetics; HapMap; GWAS; Genome-wide threshold; autocorrelation; linkage disequilibrium
• ISSN: 0003-4800; 1469-1809; 1469-1809
• DOI: 10.1111/ahg.12095

Summary The number of effectively independent tests performed in genome‐wide association studies (GWAS) varies by population, making a universal P‐value threshold inappropriate. We estimated the number of independent SNPs in Phase 3 HapMap samples by: (1) the LD‐pruning function in PLINK, and (2) an autocorrelation‐based approach. Autocorrelation was also used to estimate the number of independent SNPs in whole genome sequences from 1000 Genomes. Both approaches yielded consistent estimates of numbers of independent SNPs, which were used to calculate new population‐specific thresholds for genome‐wide significance. African populations had the most stringent thresholds (1.49 × 10−7 for YRI at r2 = 0.3), East Asian populations the least (3.75 × 10−7 for JPT at r2 = 0.3). We also assessed how using population‐specific significance thresholds compared to using a single multiple testing threshold at the conventional 5 × 10−8 cutoff. Applied to a previously published GWAS of melanoma in Caucasians, our approach identified two additional genes, both previously associated with the phenotype. In a Chinese breast cancer GWAS, our approach identified 48 additional genes, 19 of which were in or near genes previously associated with the phenotype. We conclude that the conventional genome‐wide significance threshold generates an excess of Type 2 errors, particularly in GWAS performed on more recently founded populations.