Analysis of untyped SNPs: maximum likelihood and imputation methods

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Untyped single nucleotide polymorphisms (SNPs) analysis is crucial for localizing disease variants and enabling meta-analyses across different genotyping platforms.
• Leveraging linkage disequilibrium (LD) from external reference panels can infer unknown SNP genotypes from observed ones.

Purpose of the Study:

• To present and compare two distinct computational approaches for inferring untyped SNP genotypes.
• To evaluate the performance of maximum-likelihood and imputation methods in genetic association studies.

Main Methods:

• Developed a maximum-likelihood approach integrating genotype prediction and association parameter estimation.
• Utilized a two-stage imputation approach, first predicting untyped genotypes then performing downstream association analysis.
• Conducted extensive simulations to compare bias, type I error, power, and confidence interval coverage.

Main Results:

• The maximum-likelihood approach provides consistent and efficient estimators for genetic effects and gene-environment interactions with accurate variance estimation.
• The imputation approach controls type I error for single-SNP tests but may not properly control it for multiple-SNP effects or gene-environment interactions.
• Imputation generally results in biased genetic effect estimators and underestimated variances.

Conclusions:

• The maximum-likelihood method is superior for accurate estimation of genetic effects and interactions compared to standard imputation.
• Both methods were illustrated using genome-wide data from the Wellcome Trust Case-Control Consortium.