LTBT implements a novel rare-variant association test for binary traits that jointly models the contribution of rare variants and PRS under a liability threshold model. We define the total liability as the additive sum of rare-variant effect (γ), PRS, and unexplained risk factors (Figure A). Under a liability threshold model, the disease outcome is determined by whether an individual’s total liability is over the threshold determined by the prevalence of disease in population. By modeling the conditional distribution of PRS given the rare variant effect size γ (Figure B), we compare the likelihood of observed data under the null (γ=0) and alternative (γ≠0) hypotheses.
For inquiries on LTBT, please contact Sung Chun (sung.g.chun@gmail.com).
LTBT is implemented as an R module. Please download and install the LTBT R module as below:
R CMD install ltbt_1.0.tar.gzWe provide a sample test dataset RTEL1.tsv and test code test.r.
RTEL1.tsv contains deidentified summary-level data for the gene Regulator of Telomere Elongation Helicase 1 (RTEL1), a known familial pulmonary fibrosis gene. The file contains PRS values for 136 IPF cases and 230,496 healthy controls (Figure C). Of the 136 cases, 3 subjects carry a rare predicted loss-of-function (pLoF) variant in RTEL1. In comparison, 127 control subjects carry such a rare pLoF variant. The PRS value is pre-standardized to the mean of 0 and variance of 1 in an unascertained population of European ancestry. PRS was calculated by aggregating risk allele dosages weighted by their GWAS effect over 14 GWAS SNPs. In Figure C, black dots indicate PRS values for RTEL1 pLoF carriers. Violin plots represent the overall distribution of PRS among non-carriers. Note the average PRS of pLoF-carrying patients is closer to 0 compared to non-carrier patients, suggesting the former does not need a strong disease-predisposing PRS to manifest disease.
LTBT scans the likelihood space in a grid search. We recommend the search space of rare-variant effect (gamma.seq) as shown in test.r below. LTBT test (run.ltbt()) requires the genotypes, polygenic risk scores (PRS), outcomes for all subjects in a cohort, along with the disease prevalence in an unascertained population (0.005 in this case). The genotypes are encoded as 0 or 1, respectively, depending on whether an individual carries a functional rare variant or not. The PRS should be standardized to the mean of 0 and variance of 1 in an unascertained population. The outcomes are represented as 0 or 1 for controls and cases, respectively.
library(ltbt)
gamma.seq <- seq(-5, 5, 0.05)
exdata <- read.delim("RTEL1.tsv", sep="\t")
res <- run.ltbt(gamma.seq, exdata$gt, exdata$prs, exdata$outcome, 0.005)
res$h2Lx
# 0.1266585
res$pvalue
# 4.824949e-05
res$gamma.mle
# 1.45
plot(res$gamma, res$logL, ty="l", xlab="gamma", ylab="Log Likelihood Ratio")
abline(v=res$gamma.mle, col="red", lty=2)LTBT returns the estimated heritability explained by PRS (h2Lx), p-value of nested likelihood ratio test (pvalue), and most likely effect size of rare variants (gamma.mle). The curve of log likelihood ratio can be plotted as shown below:
We provide QC and data processing pipelines used in the IPF study (Chun et al.)
-
Discovery cohort (CGS-PF + UK Biobank)
- Allen2019.14SNPs.qctool.txt: 14-SNP PRS model for QCTools (Genome assembly: GRCh37).
- variant_qc.py: Hail script for Whole-Exome Sequence variant QC for CGS-PF and UK Biobank based on DNA Nexus cloud platform.
- data-cleanup.cgspf-ukb.R: Data harmonization codes for CGS-PF cases and UK Biobank controls.
-
Validation cohort (All of Us)
- Allen2019.13SNPs.b38.qctool.txt: 13-SNP PRS model for QCTools (Genome assembly: GRCh38). One SNP in 14-SNP PRS model (rs2077551) is missing in All of Us ACAF variant set.
- Calculate_PRS.py: Hail-based PRS calculation script for All of Us Workbench cloud platform.
- data-cleanup.allofus.py: Case/control definition script for All of Us Workbench cloud platform.
Sung Chun*, Ahmad Samiei, Lauren Flynn, Heidi Makrynioti, Matthew Moll, Anna L. Peljto, David A. Schwartz, Michael H Cho, Shamil R Sunyaev, Ivan O Rosas, Gary M. Hunninghake, and Benjamin A Raby*. A new liability-based rare-variant burden test identifies a novel genetic association between HTRA3 and idiopathic pulmonary fibrosis. (Preprint will be available soon)
- Joint correspondence.