Expression Quantitative Trait Loci (eQTLs) have often been used to nominate candidate genes from Genome-wide association studies (GWAS). However, commonly used methods are susceptible to false positives largely due to Linkage Disequilibrium (LD) of eQTLs with causal variants acting on the phenotype directly.

Our method, "causal-TWAS" (cTWAS), addresses this challenge by borrowing ideas from statistical fine-mapping. It is a generalization of methods for transcriptome-wide association studies (TWAS), but when analyzing any gene, it adjusts for other nearby genes and all nearby genetic variants.

While the published paper analyzes a single eQTL dataset, the latest version of cTWAS extends the method to integrate multiple groups of prediction models, allowing for joint analysis of multiple types of molecular traits, across potentially different tissues, cell types or conditions.

Use "remotes" to install the latest version of ctwas from GitHub:

install.packages("remotes")
remotes::install_github("xinhe-lab/ctwas",ref = "multigroup_test")

Currently, ctwas has only been tested on Linux systems.

We recommend installing and running ctwas on a high-performance computing system.

Running a cTWAS analysis involves four main steps:

1. Preparing the input data.

2. Computing the gene z-scores.

3. Estimating the model parameters.

4. Fine-mapping.

The outputs of cTWAS are posterior inclusion probabilities (PIPs) for all variants and molecular traits.

To learn more about the ctwas R package, we recommend starting with this introductory tutorial:

A minimal tutorial of how to run cTWAS without LD

To run the full cTWAS, a few more tutorials including:

• Preparing input data

• Running cTWAS main function

• Running cTWAS by modules

• Summarizing and visualizing cTWAS results

• Post-processing cTWAS results

You can browse source code and report a bug here.

If you find the ctwas package or any of the source code in this repository useful for your work, please cite:

Zhao S, Crouse W, Qian S, Luo K, Stephens M, He X. Adjusting for genetic confounders in transcriptome-wide association studies improves discovery of risk genes of complex traits. Nature Genetics 56, 336–347 (2024). https://doi.org/10.1038/s41588-023-01648-9

We have pre-computed the LD matrices of European samples from UK Biobank. They can be downloaded here.

We have the lists of reference variant information from all the LD matrices in the genome in hg38 and hg19.

cTWAS requires the expression prediction models, or weights, of genes. The pre-computed weights of GTEx expression and splicing traits can be downloaded from PredictDB.

We thank the authors of susieR package for using their codes.

Original susieR code obtained by:

git clone [email protected]:stephenslab/susieR.git
git checkout c7934c0

Minor edits to make it accept different prior variances for each variable.