In marf-at-vt/GWAS.BAYES: Bayesian analysis of Gaussian GWAS data

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  tidy = FALSE
)

library(GWAS.BAYES)

Introduction

The GWAS.BAYES package provides statistical tools for the analysis of Gaussian GWAS data. Currently, GWAS.BAYES contains functions to perform BICOSS which is a novel iterative two step Bayesian procedure [@BICOSS] that, when compared to single marker analysis (SMA), increases the recall of true causal SNPs and drastically reduces the rate of false discoveries. Full details on the BICOSS procedure can be found in @BICOSS.

This vignette shows an example of how to use the BICOSS function provided in GWAS.BAYES to analyze GWAS data. Data has been simulated under a linear mixed model from 9,000 SNPs for 328 A. Thaliana ecotypes. The GWAS.BAYES package includes as R objects the 9,000 SNPs, the simulated phenotypes, and the kinship matrix used to simulate the data.

Functions

The function implemented in GWAS.BAYES is described below:

BICOSS Performs BICOSS, as proposed by @BICOSS, using linear mixed models for a given numeric phenotype vector y, a SNP matrix encoded numerically SNPs, and a realized relationship matrix or kinship matrix kinship. The BICOSS function returns the indices of the SNP matrix that were identified in the best model found by the BICOSS algorithm.

Model/Model Assumptions

The model for GWAS analysis used in the GWAS.BAYES package is

\begin{equation} \textbf{y} = X \boldsymbol{\beta} + Z \textbf{u} + \boldsymbol{\epsilon} \ \text{where} \ \boldsymbol{\epsilon} \sim N(\textbf{0},\sigma^2 I) \ \text{and} \ \textbf{u} \sim N(\textbf{0},\sigma^2 \tau K), \end{equation}

where

$\textbf{y}$ is the vector of phenotype responses.
$X$ is the matrix of SNPs (single nucleotide polymorphisms).
$\boldsymbol{\beta}$ is the vector of regression coefficients that contains the effects of the SNPs.
$Z$ is an incidence matrix relating the random effects associated with the kinship structure.
$\textbf{u}$ is a vector of random effects associated with the kinship structure to the phenotype responses.
$\boldsymbol{\epsilon}$ is the error vector.
$\sigma^2$ is the variance of the errors.
$\tau$ is a parameter related to the variance of the random effects.
$K$ is the kinship matrix.

Currently, all functions in GWAS.BAYES assume the errors of the fitted model are Gaussian. To speed up computations, GWAS.BAYES utilizes spectral decomposition techniques similar to that of @Kang1709 as well as population parameters previously determined (P3D,@P3D).

Example

The BICOSS function requires a vector of observed phenotypes, a matrix of SNPs, and a kinship matrix First, the vector of observed phenotypes must be a numeric vector or a numeric $n \times 1$ matrix. GWAS.BAYES does not allow the analysis of multiple phenotypes at the same time. In this example, the vector of observed phenotypes was simulated from a linear mixed model. Here are the first five elements of the simulated vector of phenotypes:

Y[1:5]

Second, the SNP matrix has to contain numeric values where each column corresponds to a SNP of interest and the $i$th row corresponds to the $i$th observed phenotype. In this example, the SNPs are a subset of the TAIR9 genotype dataset and all SNPs have minor allele frequency greater than 0.01. Here are the first five rows and five columns of the SNP matrix:

SNPs[1:5,1:5]

Third, the kinship matrix is an $n \times n$ positive semi-definite matrix containing only numeric values. The $i$th row or $i$th column quantifies how observation $i$ is related to other observations. Here are the first five rows and five columns of the kinship matrix:

kinship[1:5,1:5]

BICOSS

The function BICOSS implements the BICOSS method for linear mixed models with Gaussian errors. This function takes as inputs the observed phenotypes, the SNPs coded numerically, the kinship matrix, and whether or not to use the P3D approach. Further, the other inputs of BICOSS are the FDR nominal level, the maximum number of iterations of the genetic algorithm in the model selection step, and the number of consecutive iterations of the genetic algorithm with the same best model for convergence. The full details of BICOSS are available in @BICOSS. The default values of maximum iterations and the number of iterations are the values used in the simulation study in @BICOSS, that is, 400 and 40 respectively.

Here we illustrate the use of BICOSS with a nominal FDR of 0.05 and with the P3D approach in both the screening and the model selection steps.

BICOSS_P3D <- BICOSS(Y = Y, SNPs = SNPs, 
                     kinship = kinship,FDR_Nominal = 0.05,P3D = TRUE,
                     maxiterations = 400,runs_til_stop = 40)
BICOSS_P3D$best_model

BICOSS outputs the best model in a named list. The best model values correspond to the indices of the SNP matrix. Further, estimating the variance components for each model in the screening and model selection steps is possible by setting P3D = FALSE. This is a much slower option.

BICOSS_Exact <- BICOSS(Y = Y, SNPs = SNPs, 
                       kinship = kinship,FDR_Nominal = 0.05,P3D = FALSE,
                       maxiterations = 400,runs_til_stop = 40)
BICOSS_Exact$best_model

As expected, using P3D or not using P3D leads to slightly different sets of identified SNPs. Because this is simulated data, we can compute the number of true positives and the number of false positives.

## The true causal SNPs in this example are
True_Causal_SNPs <- c(450,1350,2250,3150,4050,4950,5850,6750,7650,8550)
## Thus, the number of true positives is
sum(BICOSS_P3D$best_model %in% True_Causal_SNPs)
## The number of false positives is
sum(!(BICOSS_P3D$best_model %in% True_Causal_SNPs))