Deep learning-based phenotype imputation on population-scale biobank data increases genetic discoveries.

Area of Science:

• Genetics
• Bioinformatics
• Artificial Intelligence

Background:

• Biobanks are crucial for human genetics research, housing deep phenotypic and genomic data.
• Missing phenotypic data in biobanks limits their utility for genetic studies.
• Existing imputation methods struggle with extensive missingness in large-scale datasets.

Purpose of the Study:

• To introduce AutoComplete, a novel deep learning imputation method for missing phenotypes.
• To evaluate AutoComplete's performance on population-scale biobank data, specifically the UK Biobank.
• To demonstrate the utility of deep learning for enhancing genetic discovery through phenotype imputation.

Main Methods:

• Developed AutoComplete, a deep learning model for phenotype imputation.
• Applied AutoComplete to phenotype data from approximately 300,000 individuals in the UK Biobank.
• Compared imputation accuracy and genetic similarity of imputed phenotypes against existing methods.

Main Results:

• AutoComplete significantly improved imputation accuracy compared to existing methods.
• Imputed phenotypes were genetically similar to observed phenotypes across three traits.
• The method effectively doubled the average sample size for genetic analyses.
• Genome-wide association studies using imputed phenotypes revealed a greater number of associated loci.

Conclusions:

• Deep learning-based phenotype imputation, as demonstrated by AutoComplete, is effective for biobank data.
• AutoComplete enhances the utility of existing biobank resources for genetic research.
• This approach increases statistical power, facilitating more robust genetic discoveries.