Evaluation of Sex-Aware PrediXcan Models for Predicting Gene Expression

Area of Science:

• Genetics
• Systems Biology
• Bioinformatics

Background:

• Gene-based methods like PrediXcan predict gene expression using genetic data.
• Sex differences in gene expression and genetic architecture are known but not incorporated into current models.

Purpose of the Study:

• To develop and evaluate sex-aware PrediXcan models for improved gene expression prediction.
• To investigate the impact of sex-specific genetic architecture on gene expression.

Main Methods:

• Built sex-aware PrediXcan models using whole blood transcriptomic data from GTEx.
• Validated models using lymphoblast RNA sequencing data from the 1000 Genomes Project.
• Evaluated prediction performance (R2) for autosomal genes in males and females.

Main Results:

• Successfully predicted 1,149 genes in males and 623 in females; 3,511 genes were not sex-specific.
• Sex-specific models showed improved prediction for 15% of sex-specific genes.
• Several sex-specific genes, including those in mitochondrial metabolism, had significantly better prediction with sex-specific weights.

Conclusions:

• Sex-aware PrediXcan models provide robust sex-specific prediction signals.
• Genetic architecture contributes minimally to sex-specific expression, but sex-specific models offer utility.
• Future research should explore X chromosome and tissue specificity for genetically regulated expression.