Simulation Research And Comparison Of Collapsing Methods And Non-collapsing Methods For Rare Variants Correlation Analysis

Purpose Complex disease is the result of genetic and environmental factors. Toidentify diseases with complex genetic variation, genetic epidemiologists have proposed avariety of statistical methods based on genetic linkage and/or genetic association analysis.In recent years, genome-wide association studies (GWAS) has made great achievements inthe common variants associated with complex diseases or traits analysis. However, thecommon variants explain only a small proportion of the heritability of the disease or trait.This phenomenon is known as "Missing heritability." A likely reason lies in the strength ofthe presence of rare variants for complex diseases. However, because of the low frequencyof rare variants in the population, it will be ineffective if still using statistical methods toanalyze common variants in GWAS identification of rare variants on disease impact. Withthe advances in whole-genome sequencing technology, data on rare variants have becomeincreasingly available, and many investigators hope that rare variants will enhance ourunderstanding of the biological mechanisms of human diseases and traits. When analysisassociation of rare variants and disease，our study could provide reference and theoreticalbasis for molecular genetic epidemiology to choice statistical methods appropriately, andcould provide supports Theoretically and methodologically for practical applicationpromotion and popularization of rare variants association test methods. Methods According to effect pattern hypothesis of DNA sequence variants to diseasestates，study10rare variants association test methods, by changing the sample size,number of non-causal rare variants，effect size and its directions of causal rare variants,causal rare variants’ weight and the Linkage disequilibrium(LD) state levels in all rarevariants, systematically simulate a variety of genetic contexts of different combinations offactors, investigate appropriate conditions for different rare variants association testsprinciple modes, and compare and contrast type I error and power between them indifferent genetic contexts. All methods computing processes are carried out by R software(version3.0.2).Results All methods’ type I error were at the level of around0.05. When all the causalrare variants’ effect sizes and its directions were the same, as the sample size and LDparameters increased and the number of non-causal rare variants decreased, the power of allmethods gradually increased; When at small and medium sample size and LD parameterswere0, the Performance of three non-direction considered methods (CMC、w-Sum andSUM) should be more prominent if there is no or few non-causal rare variants. When all thecausal rare variants’ effect sizes were the same but its directions were not the same, totally,with the increase of LD parameters and the number of non-causal rare variants, the powersof direction considered collapsing methods (SSU, SSU and aSUM）and non-collapsingmethods (C-α,SKAT_linear, SKAT_wlinear and RR) both higher than non-directionconsidered collapsing methods. When LD parameters were large, RR optimal performance.When causal rare variants’ weight equal to variance reciprocal of MAF，totally, as the LDparameters and the number of non-causal rare variants increase, the power of all methodsincreased. When there were no non-causal rare variants, the power of w-Sum were largerthen CMC’s. When LD parameters equal to0and0.5, the power of SSUw andSKAT_wlinear were both larger then SSU and SKAT_linear respectively.Conclusion All methods could show good performance in a harmonious geneticScenario. Non-direction considered collapsing methods are suitable for simple geneticenvironments, and the other methods can maintain good performance even in a morecomplex genetic environment. The two methods are complementary.