SLINGER: large-scale learning for predicting gene expression.

Area of Science:

• Genomics
• Systems Biology
• Statistical Genetics

Background:

• Single nucleotide polymorphisms (SNPs) are known to predict gene expression.
• Previous models focused on cis-neighborhood SNPs, limiting predictive scope.
• Gene expression prediction from SNPs has shown associations with disease status.

Purpose of the Study:

• To develop a novel predictive approach, SLINGER, that removes the cis-neighborhood constraint.
• To evaluate the performance of genome-wide SNPs versus cis-neighborhood SNPs for gene expression prediction.
• To assess the impact of SLINGER on the fidelity of disease associations in genome-wide association studies.

Main Methods:

• Proposed SLINGER, a penalized linear model utilizing genome-wide SNPs.
• Compared SLINGER's performance against cis-neighborhood-only SNP models.
• Applied SLINGER to seven Wellcome Trust Case-Control Consortium genome-wide association studies.

Main Results:

• SLINGER predicts expression for more genes compared to cis-neighborhood-only models.
• The new models significantly improve prediction accuracy for numerous genes.
• SLINGER maintains a comparable number of features to cis-only models due to penalized regression.
• SLINGER models show greater fidelity to actual gene expression values in association studies.

Conclusions:

• Genome-wide SNPs enhance gene expression prediction accuracy and scope.
• SLINGER offers a more powerful approach for genetic prediction of gene expression.
• SLINGER improves the reliability of genetic association studies by linking genotype to gene expression with higher fidelity.