Combinatorial and statistical prediction of gene expression from haplotype sequence

Area of Science:

• Genomics
• Computational Biology
• Statistical Genetics

Background:

• Genome-wide association studies (GWAS) identify genetic variants linked to diseases.
• Expression quantitative trait loci (eQTL) studies link variants to gene expression, but prediction accuracy remains a challenge.
• Accurate gene expression prediction is crucial for post hoc GWAS analysis and prioritizing genes.

Purpose of the Study:

• To develop novel algorithms for predicting gene expression from genetic data, relaxing independence and additivity assumptions.
• To introduce and formalize a gene expression classification problem.
• To compare the performance of new methods against existing state-of-the-art approaches.

Main Methods:

• Developed HAPLEXR algorithm combining haplotype clustering with allelic dosages for regression-based prediction.
• Introduced gene expression classification and developed HAPLEXD using suffix trees and spectral clustering.
• Penalized model complexity by prioritizing genetic clusters with significant expression effects.

Main Results:

• HAPLEXD achieved higher overall classification accuracy across five GTEx v8 tissues.
• HAPLEXR demonstrated superior prediction accuracy for about half of the genes tested.
• HAPLEXR selected smaller, more interpretable variant and haplotype features enriched in gene regulation annotations.

Conclusions:

• Explicitly modeling non-dosage dependent and intragenic epistatic effects is important for accurate gene expression prediction.
• The developed HAPLEXR and HAPLEXD algorithms offer improved tools for genotype-to-gene expression prediction and classification.
• These methods can enhance the interpretation of GWAS findings and gene prioritization for functional studies.