standardize: Standardize summary statistics for MRMix analysis
In gqi/MRMix: Mendelian Randomization Analysis Using Mixture Models (MRMix)

Description Usage Arguments Details Value References

View source: R/standardize.R

1) For both binary and continuous traits, this function standardizes GWAS summary statistics by genotypic variance; 2) In addition, for continuous phenotype, this function standardizes summary statistics by phenotypic variance. This function is designed for GWAS estimates from linear or logistic regression. Do not use for other models.

1	standardize(betahat_x, betahat_y, sx, sy, xtype, ytype, nx, ny, MAF)

`betahat_x`	GWAS effect estimates of the exposure. Vector of length `K`, where `K` is the number of instruments (SNPs).
`betahat_y`	GWAS effect estimates of the outcome. Vector of length `K`.
`sx`	Standard error of `betahat_x`. Vector of length `K`.
`sy`	Standard error of `betahat_y`. Vector of length `K`.
`xtype`	Is the exposure a continuous or binary trait? Set to `xtype="continuous"` or `xtype="binary"`. Or set to `xtype="n"` if exposure summary statistics do not need to be standardized.
`ytype`	Is the outcome a continuous or binary trait? Set to `ytype="continuous"` or `ytype="binary"`. Or set to `ytype="n"` if outcome summary statistics do not need to be standardized.
`nx`	SNP-specific sample size (recommended) or total sample size of the study associated with the exposure. Vector of length `K` or a single number. Set to `NULL` if trait is binary. Summary statistics for binary traits are standardized by the genotypic variance which can be calculated using the minor allele frequency (MAF) under Hardy-Weinberg equilibrium. Hence sample size is not needed for binary traits.
`ny`	SNP-specific sample size (recommended) or total sample size of the study associated with the outcome. Vector of length `K` or a single number. Set to `NULL` if trait is binary for the same reason as for `nx`.
`MAF`	Minor allele frequency. Vector of length `K`. Set to `NULL` if both traits are continuous. Summary statistics for continuous traits are standardized as z statistics rescaled by sample size, hence MAF is not needed.

Using the exposure X as an example: 1) For continuous phenotypes analyzed with linear regression, data are standardized by betahat_x_std=betahat_x/(sx*sqrt(nx)); sx_std=1/sqrt(nx). Note that this standardization assumes that GWAS was conducted without covariate adjustment or the covariates do not have strong effects on Y. If the covariates have strong effects on Y, set nx equal to the effective sample size, which can be approximated by N/(1-R2), where N is the sample size associated with the study for X and R2 is the R-squared for the covariates. 2) For binary phenotypes analyzed with logistic regression, data are standardized by betahat_x_std=betahat_x*sqrt(2*MAF*(1-MAF)); sx_std=sx*sqrt(2*MAF*(1-MAF)). Same formulas apply to the outcome Y.

A list that contains

`betahat_x_std`	Standardized `betahat` for the exposure.
`betahat_y_std`	Standardized `betahat` for the outcome.
`sx_std`	Standard error of `betahat_x_std`.
`sy_std`	Standard error of `betahat_y_std`.

1. Qi, Guanghao, and Nilanjan Chatterjee. "Mendelian randomization analysis using mixture models for robust and efficient estimation of causal effects." Nature Communications 10.1 (2019): 1941.

2. Qi, Guanghao, and Nilanjan Chatterjee. "A Comprehensive Evaluation of Methods for Mendelian Randomization Using Realistic Simulations of Genome-wide Association Studies." bioRxiv (2019): 702787.

gqi/MRMix documentation built on Jan. 30, 2020, 1:35 a.m.