Towards Faster Gene Expression Prediction via Dimensionality Reduction and Feature Selection.

Area of Science:

• Genomics
• Computational Biology
• Statistical Genetics

Background:

• Gene expression is influenced by genetic factors.
• Elastic nets are effective for predicting gene expression from genotype data.
• Reducing genetic data dimensionality is crucial for computational efficiency.

Purpose of the Study:

• To evaluate dimensionality reduction techniques for gene expression prediction models.
• To assess the impact of Principal Component Analysis (PCA) and linkage disequilibrium (LD) pruning on elastic net model performance and computation time.

Main Methods:

• Applied PCA and LD pruning to genetic variants.
• Used elastic net models to predict gene expression from genotype data.
• Compared model performance (R-squared) and computational time with different input data reduction strategies.

Main Results:

• Elastic nets perform best with all genetic variants, but PCA significantly reduces computation time.
• 100 principal components reduced computation time by over 80% with only an 8% loss in R-squared.
• LD pruning was not effective for reducing genetic variants for gene expression prediction.

Conclusions:

• PCA is an effective method for reducing the computational burden of gene expression prediction models.
• This approach is particularly valuable given the scale of genomic datasets (27,000+ genes, 50+ tissues).
• PCA offers a practical solution for accelerating large-scale gene expression analyses.