A-TWAS: An aggregated transcriptome-wide association study model incorporating multiple Bayesian priors

Motivation Transcriptome-wide association study (TWAS) is a significant methodology utilized for identifying associations between genes and diseases by integrating transcriptome and genome-wide association studies (GWAS) data. The approach has been successful in pinpointing risk genes for various diseases, including Alzheimer’s disease, schizophrenia, and different types of cancers. TWAS typically involves two key steps: imputation and association analysis. In the imputation step, the original PrediXcan employs the elastic-net model, while subsequent research endeavors have delved into more intricate models such as FUSION and TIGAR. Despite these advancements, the existing individual methods may not capture the intricate genotype-expression relationships in a comprehensive way. Given the complexity of genetic contributions of genotypes to gene expression, sophisticated modeling techniques are imperative. In response to this, we have introduced Aggregated-TWAS (A-TWAS), a comprehensive tool that amalgamates multiple imputation models to accommodate diverse genetic architectures. A-TWAS utilizes several continuous shrinkage priors, including Laplace, Horseshoe, and Horseshoe+, to ensure precise imputation results while incorporating the benefits of Bayesian variable selection methods. In the association phase of TWAS, we employ the aggregated Cauchy association test (ACAT) to obtain an omnibus p -value. Result We demonstrate the effectiveness of A-TWAS through comprehensive simulations and real data analyses. Simulation studies highlight that A-TWAS produces substantial improvements in predictive R 2 and statistical power, while maintaining the type error I at a low level. Furthermore, TWASs are conducted on schizophrenia and obsessive-compulsive symptoms datasets. In comparison to standard methods, the Bayesian imputation model demonstrates superior accuracy, and A-TWAS, by integrating multiple Bayesian priors, identifies a notably greater number of disease-relevant genes. Availability and Implementation The source code of A-TWAS is available at https://github.com/Yilan-Liang/A-TWAS .