Transferability of Single- and Cross-Tissue Transcriptome Imputation Models Across Ancestry Groups

Area of Science:

• Genetics
• Bioinformatics
• Population Genetics

Background:

• Transcriptome-wide association studies (TWAS) link genetically regulated gene expression to complex traits.
• TWAS accuracy depends on imputation models using expression quantitative trait loci (eQTLs).
• Existing resources like GTEx are ancestry-biased, potentially limiting TWAS in diverse populations.

Purpose of the Study:

• To evaluate the transferability of gene expression imputation models across different ancestry groups.
• To compare the performance of tissue-specific (PrediXcan) and cross-tissue (UTMOST) imputation models.
• To assess the impact of ancestry and tissue context on eQTL detection.

Main Methods:

• Trained PrediXcan and UTMOST models on European ancestry data from GTEx.
• Tested model performance on independent European ancestry and African American datasets.
• Compared imputation accuracy between models and across ancestry groups.

Main Results:

• Both imputation models performed best when trained and tested on the same ancestry group.
• The cross-tissue UTMOST model generally outperformed the tissue-specific PrediXcan model.
• eQTL detection accuracy is significantly influenced by both ancestry and tissue type.

Conclusions:

• Gene expression imputation accuracy varies across ancestry groups.
• Developing ancestry-specific reference panels is crucial for improving TWAS.
• Addressing ancestry bias in genetic studies enhances understanding of complex trait biology.