Evaluation of logistic Bayesian LASSO for identifying association with rare haplotypes

Abstract

It has been hypothesized that rare variants may hold the key to unraveling the genetic transmission mechanism of many common complex traits. Currently, there is a dearth of statistical methods that are powerful enough to detect association with rare haplotypes. One of the recently proposed methods is logistic Bayesian LASSO for case-control data. By penalizing the regression coefficients through appropriate priors, logistic Bayesian LASSO weeds out the unassociated haplotypes, making it possible for the associated rare haplotypes to be detected with higher powers. We used the Genetic Analysis Workshop 18 simulated data to evaluate the behavior of logistic Bayesian LASSO in terms of its power and type I error under a complex disease model. We obtained knowledge of the simulation model, including the locations of the functional variants, and we chose to focus on two genomic regions in the MAP4 gene on chromosome 3. The sample size was 142 individuals and there were 200 replicates.Despite the small sample size, logistic Bayesian LASSO showed high power to detect two haplotypes containing functional variants in these regions while maintaining low type I errors. At the same time, a commonly used approach for haplotype association implemented in the software hapassoc failed to converge because of the presence of rare haplotypes. Thus, we conclude that logistic Bayesian LASSO can play an important role in the search for rare haplotypes.

Comparison of haplotype-based statistical tests for disease association with rare and common variants

“…On the other hand, BhGLM is highly conservative with almost all type I error rates close to 0. LBL, with the cutoff of 2, has controlled type I error rates, each below 5%, consistent with previous studies [12,14,15,26]. Similar trend is seen for other sample sizes (see Supplementary Tables S1 and S2).…”

“…The phenotypes of systolic and diastolic blood pressures were generated [25,31]. We used a binary hypertension status derived from these phenotypes and medication status to define case/control status as in [26]. The original data are on 20 extended families, so founders were extracted from these families to get a sample of 142 independent individuals.…”

“…As this is a small sample size especially for detecting rare variants, following [26], we focus on analysis of two genomic regions, each of 8000 base pairs (bp) and centered at the following two functional (causal) variants: (1) the most frequent variant at location 47 957 741 bp with minor allele frequency (MAF) ¼ 0.378 (Region 1) and (2) the variant with the strongest effect size at location 48 040 283 bp with MAF ¼ 0.032 (Region 2). Both of these regions are in MAP4 gene, which has a total of 15 functional variants, with MAF ranging from 0.0016 to 0.378.…”

“…Out of them, five are in Region 1 and two are in Region 2. As in [26], we included SNPs with MAF > 0.01 in forming the two haplotype blocks in these regions, thus requiring at least two copies of minor allele of a SNP to be present in a sample of size 142. In Simulation results for individual haplotype-specific methods in RC scenario with n ¼ 1000.…”

“…However, we do not intend to provide an in-depth comparison of haplotyping versus collapsing approaches, which has been reported elsewhere [11]. We use our own simulated data as well as Genetic Analysis Workshop (GAW) 18 simulated data [25,26]. We further apply the methods to two real data sets-hypertension data from GAW18 and Dallas Heart Study (DHS) case-control data [27,28].…”