Bias Characterization in Probabilistic Genotype Data and Improved Signal Detection with Multiple Imputation

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Missing data are prevalent in genetic studies due to varying technologies.
• Unaddressed missingness introduces bias in genome-wide association studies (GWAS).
• Current imputation methods may not fully account for genotype uncertainty.

Purpose of the Study:

• To systematically evaluate existing methods for analyzing imputed genetic data.
• To assess the performance of Multiple Imputation (MI) in handling genotype uncertainty.
• To characterize biases arising from imputation and association analysis.

Main Methods:

• Comparative analysis of existing imputed data analysis techniques.
• Evaluation of Multiple Imputation (MI) for genetic association studies.
• Bias characterization at both imputation and association levels.

Main Results:

• Bias is introduced by inconsistent genotype probabilities from imputation algorithms.
• Inadequate modeling of genotype uncertainty in association analysis also causes bias.
• Multiple Imputation (MI) performs comparably or superiorly to existing methods.

Conclusions:

• Multiple Imputation (MI) offers a robust framework for analyzing genetic data with missingness.
• MI effectively models uncertainty, reducing bias in genome-wide association studies.
• This paradigm facilitates adaptation of existing methods for uncertain genotype data.