Evaluation of Genotype-Based Gene Expression Model Performance: A Cross-Framework and Cross-Dataset Study

Area of Science:

• Genetics
• Bioinformatics
• Neuroscience

Background:

• Predicting gene expression from genotype data is crucial for studying tissues like the brain.
• Existing methods like eGenScore and PrediXcan aim to predict gene expression using genetic information.

Purpose of the Study:

• To compare the performance of eGenScore and PrediXcan for predicting gene expression in the frontal cortex.
• To evaluate the impact of different training datasets (brain-specific vs. multi-tissue) on prediction accuracy.

Main Methods:

• Development and comparison of eGenScore (polygenic/poly-variation) and PrediXcan (regularized linear regression with elastic nets).
• Training and validation of expression quantitative trait loci (eQTL) models using frontal cortex data.
• Internal cross-validation and external validation using the CommonMind Consortium database.

Main Results:

• PrediXcan demonstrated superior performance over eGenScore across all tested training datasets.
• eQTL models trained with GTEx (multi-tissue) data showed higher prediction accuracy in the frontal cortex compared to those trained with BrainEAC (brain-specific) data when using PrediXcan.

Conclusions:

• PrediXcan is a more effective method for predicting gene expression in brain tissues compared to eGenScore.
• Multi-tissue gene expression datasets like GTEx can improve the accuracy of predicting gene expression in specific brain regions.