Detect Outliers Using Dentist Algorithm

DENTIST (Detecting Errors iN analyses of summary staTISTics) is a quality control tool for GWAS summary data. It uses linkage disequilibrium (LD) information from a reference panel to identify and correct problematic variants by comparing observed GWAS statistics to predicted values. It can detect errors in genotyping/imputation, allelic errors, and heterogeneity between GWAS and LD reference samples.

Usage

dentist(
  sumStat,
  R = NULL,
  X = NULL,
  nSample = NULL,
  windowSize = 2e+06,
  windowMode = c("distance", "count"),
  pValueThreshold = 5.0369e-08,
  propSVD = 0.4,
  gcControl = FALSE,
  nIter = 10,
  gPvalueThreshold = 0.05,
  duprThreshold = 0.99,
  ncpus = 1,
  correctChenEtAlBug = TRUE,
  minDim = 2000,
  ldMethod = "sample"
)

Arguments

sumStat: A data frame containing summary statistics, including 'pos' or 'position' and 'z' or 'zscore' columns.
R: Square LD correlation matrix. Provide either R or X.
X: Genotype matrix (samples x SNPs). If provided, LD is computed via computeLd(X) and nSample defaults to nrow(X).
nSample: The number of samples in the LD reference panel (NOT the GWAS sample size). This controls the SVD truncation rank K = min(idx_size, nSample) * propSVD. Required when R is provided; inferred from X when X is provided.
windowSize: The size of the window for dividing the genomic region in distance mode (base pairs). Default is 2000000 (2 Mb). Only used when windowMode = "distance".
windowMode: Character string specifying the windowing strategy: "distance" (default) creates windows by physical distance using segmentByDist (C++ --wind-dist), and "count" creates windows by variant count using segmentByCount (C++ --wind).
pValueThreshold: The p-value threshold for significance. Default is 5e-8.
propSVD: The proportion of singular value decomposition (SVD) to use. Default is 0.4.
gcControl: Logical indicating whether genomic control should be applied. Default is FALSE.
nIter: The number of iterations for the Dentist algorithm. Default is 10.
gPvalueThreshold: The genomic p-value threshold for significance. Default is 0.05.
duprThreshold: The absolute correlation r value threshold to be considered duplicate. Default is 0.99.
ncpus: The number of CPU cores to use for parallel processing. Default is 1.
correctChenEtAlBug: Logical indicating whether to correct the Chen et al. bug. Default is TRUE.
minDim: In distance mode: minimum number of SNPs per block (default 2000). In count mode: the number of variants per window (i.e., the window size).
ldMethod: Character string specifying the LD computation method when X is provided. Passed to computeLd. One of "sample" (default), "population", or "gcta". Ignored when R is provided directly.

Value

A data frame containing the imputed result and detected outliers.

The returned data frame includes the following columns:

original_z: The original z-score values from the input sumStat.
imputed_z: The imputed z-score values computed by the Dentist algorithm.
rsq: The coefficient of determination (R-squared) between original and imputed z-scores.
iter_to_correct: The number of iterations required to correct the z-scores, if applicable.
index_within_window: The index of the observation within the window.
index_global: The global index of the observation.
outlier_stat: The computed statistical value based on the original and imputed z-scores and R-squared.
outlier: A logical indicator specifying whether the observation is identified as an outlier based on the statistical test.

Details

Windowing supports two modes matching the original DENTIST C++ binary:

"distance" (default): Uses the segmentingByDist algorithm (C++ --wind-dist), implemented in segmentByDist. Windows span a fixed physical distance (windowSize bp).
"count": Uses the segmentedQCed algorithm (C++ --wind), implemented in segmentByCount. Windows contain a fixed number of variants (minDim). Useful when regions have sparse variants where distance-based windows would create windows with too few variants.

The correctChenEtAlBug parameter affects the iterative filtering in two ways:

Comparison between iteration index t and nIter (explained in source code)
The !grouping_tmp operator bug (explained in source code)

Examples

# Example usage of dentist
dentist(sumStat, R = ldMat, nSample = nSample)
#> Error: object 'ldMat' not found