effectSizeEstimation: Estimate Non-Linear Model for the Effect of Genotype on Gene...

In ACMEeqtl: Estimation of Interpretable eQTL Effect Sizes Using a Log of Linear Model

Description

Estimate non-linear model for the effect of genotype on the phenotype of interest, i.e. gene expression.

Usage

effectSizeEstimationC(x, y, cvrt)
effectSizeEstimationR(x, y, cvrt)

Arguments

x	Genotype vector. Typically having 0/1/2 values.
y	Phenotype vector. Typically gene expression in normalized raw counts.
cvrt	Matrix of covariates.

Details

The function has two implementations, one fully coded in R and a faster version with core coded in C.

Value

Returns a vector with estimated parameters and diagnostics information, such as number of iterations till convergence.

The items of the vector include:

beta0	The constant parameter in the non-linear model.
beta1	The effect size parameter in the non-linear model.
nits	Number of iterations till convergence of the estimation algorithm.
SSE	Sum of squared residuals of the fitted model.
SST	Sum of squared residuals of the model with zero effect.
F	The F test for the significance of the genotype effect.
eta	The effect size parameter for simplified model (beta1/beta0).
SE_eta	Standard error of the eta estimate.

Author(s)

Andrey A Shabalin [e-mail], John Palowitch

References

The manuscript is available at: http://onlinelibrary.wiley.com/doi/10.1111/biom.12810/full

See Also

For package overview and code examples see the package vignette via:
browseVignettes("ACMEeqtl")
or
RShowDoc("doc/ACMEeqtl.html", "html", "ACMEeqtl")

For fast testing of all local gene-SNP pairs (local eQTL) see multithreadACME.

Examples

# Model parameters
beta0 = 10000
beta = 50000

# Data dimensions
n = 1000
p = 19

# Standard deviation of covariate effects and noise
cvrtsd = 10
noisesd = 1

### Data generation
### Zero average covariates
cvrt = matrix(rnorm(n * p, sd = cvrtsd), n, p)
cvrt = t(t(cvrt) - colMeans(cvrt))

c_eff = rnorm(p, sd = cvrtsd)
error = rnorm(n, sd = noisesd)

# Generate SNPs
x = rbinom(n, size = 2, prob = 0.2)
y = log(beta0 + beta * x) + cvrt %*% c_eff + error

### Model estimation

z1 = effectSizeEstimationR(x, y, cvrt)
z2 = effectSizeEstimationC(x, y, cvrt)

### Compare the estimates

show(cbind(z1, z2))

ACMEeqtl documentation built on May 2, 2019, 4:03 p.m.