ctwas paper related analyses

Introduction

This repository contains code and data resources to accompany our research paper:

Siming, Z., Wesley, C., Sheng, Q., Kaixuan, L., Stephens, M. & Xin, H. (2022). Adjusting for genetic confounders in transcriptome-wide association studies leads to reliable detection of causal genes application to genetic fine mapping. *bioRxiv https://doi.org/10.1101/2022.09.27.509700

The ctwas R package can be accessed here. The main analyses in the paper was obtained using ctwas v0.1.29.

Perform a test simulation

An example simulation pipepline is shown as code/simulation/Snakefile-test. It was written in the snakemake workflow language.

Variables:

• codedir: Directory containing the code for simulations
• pgenfs: Path to the genotype data
• ld_R_dir: Directory containing the reference LD
• weight: Path to the weight file
• outputdir: Directory where output files will be saved
• runtag: A tag to identify the simulation
• configtags: Tags for different configuration settings
• simutags: Tags for different simulation settings

Rules:

• rule all: This is the default rule that specifies all the outputs that the workflow is expected to generate
• rule simu_prep_param: Prepares parameter files for simulations
• rule impute_expr: Runs an R script to impute expression data
• rule simulate_phenotype: Simulates phenotypes based on the imputed expression data, genotype data, and simulation parameters
• rule gwas_snp: Performs GWAS
• rule ctwas_rss_LDR: Perform causal-TWAS analysis with reference LD data

Execution Flow：

• The workflow begins by preparing simulation parameters
• It then proceeds to impute expression data, followed by phenotype simulation
• Next, it performs GWAS analyses
• Finally, it runs causal-TWAS analyses

To run the simulation:

• Go to the the code/simulation/ folder
• Run snakemake -s Snakefile-test -j 999

Simulation performed in the paper.

Input data for simulation

• Genotype data.

The individual genotype data we used is from UKB biobank, randomly selecting 80000 or 200000 samples. We then filtered samples. The columns selected for filtering samples are as follows:

- sex (31)
- UK Biobank assessment centre (54)
- age (21022)
- genetic ethnic grouping (22006)
- genetic sex (22001)
- genotype measurement batch (22000)
- missingness (22005)
- genetic PCs (22009-0.1 - 22009-0.40)
- genetic relatedness pairing (22011)
- genetic kinship (22021)
- outliers (22027)

We filtered the samples based on the following criteria:

- Remove all rows in which one or more of the values are missing, aside from the in the "outlier" and "relatedness_genetic" columns. This step removes any samples that are not marked as being "White British".
- Remove rows with mismatches between self-reported and genetic sex
- Remove "missingness" and "heterozygosity" outliers as defined by UK Biobank.
- Remove any individuals have at least one relative
- Remove any individuals that have close relatives identified from the "relatendess" calculations that weren't already identified using the kinship calculations.

This ended up with n = ~45k or ~113k samples used in our simulation. We use SNP genotype data from chr 1 to chr 22 combined from UKB. We select SNPs with minor allele frequency > 0.05 and SNPs with at least 95% genotyping rate (5% missing). There are total = 6228664 SNPs passed these filters and go into our analysis.

• LD genotype reference

We used the genotype of 2k samples from UKbiobank (randomly selected from the samples used in simulations) to serve as the LD reference.

• Expression models

The one we used in this analysis is GTEx Adipose tissue v7 dataset. This dataset contains ~ 380 samples, 8021 genes with expression, we used their lasso results. SNPs included in eQTL anlaysis are restricted to cis-locus 500kb on either side of the gene boundary. eQTLs are defined as SNPs with abs(effectize) > 1e-8 in lasso results.

Running the simulation pipeline

Simulation results in our paper comes from four pipelines, available in the code/simulation/ folder.

They are:

• Snakefile-simu_20210416, this file simulated GWAS phenotypes using 45k samples, causal effect size were simulated from normal distribution.

• Snakefile-simu_20210418, this file simulated GWAS phenotypes using 45k samples, causal effect size were simulated from normal distribution.

• Snakefile-simu_20230321, this file simulated GWAS phenotypes using 113k samples, causal effect size were simulated from normal distribution.