Multi-ancestry transcriptome prediction with functionally informed variants in TOPMed MESA improves performance of...

Area of Science:

• Genetics
• Computational Biology
• Statistical Genomics

Background:

• Accurate transcriptome prediction models are essential for multi-ancestry transcriptome-wide association studies (TWASs).
• Existing methods may not fully leverage functional information for variant selection.
• Multi-ancestry studies require robust models that perform well across diverse populations.

Purpose of the Study:

• To develop and evaluate novel transcriptome prediction models using functionally informed variants (FIVs) to enhance multi-ancestry TWAS.
• To compare the performance of FIV-based methods against the benchmark elastic net (EN) method.
• To introduce an omnibus approach for aggregating results and further improving TWAS power and gene discovery.

Main Methods:

• Trained transcriptome prediction models using RNA sequencing data from 1,287 multi-ancestry participants (TOPMed MESA) with a focus on FIVs.
• Validated model prediction accuracy on independent datasets (Geuvadis, Jackson Heart Study).
• Integrated models with multi-ancestry Genome-Wide Association Studies (GWASs) for blood cell, lipid, and pulmonary function traits; developed and tested an omnibus aggregation approach.

Main Results:

• FIV-based methods achieved comparable prediction accuracy to EN but used a smaller variant set.
• Proposed methods demonstrated improved TWAS power and accuracy (average 24% improvement over EN).
• The omnibus approach identified the most Bonferroni-significant TWAS genes across all traits and enhanced performance for blood cell traits.

Conclusions:

• Functionally informed variants significantly improve multi-ancestry transcriptome prediction models for TWAS.
• The omnibus approach offers a powerful strategy for maximizing TWAS discovery and accuracy.
• TWAS improvement is influenced by the relevance of the GWAS trait to the cell type used for transcriptome model training (PBMCs).