Rapid, Reference-Free human genotype imputation with denoising autoencoders.

Area of Science:

• Population genetics
• Genomics
• Bioinformatics

Background:

• Genotype imputation is crucial for population genetics, but standard methods face computational and privacy challenges.
• Current statistical imputation techniques struggle with accuracy in low and complex linkage disequilibrium regions.

Purpose of the Study:

• To develop and evaluate an autoencoder-based genotype imputation method.
• To overcome limitations of traditional imputation approaches, including computational cost and accuracy degradation.

Main Methods:

• An autoencoder model was trained using a large, standard reference panel covering human chromosome 22.
• The autoencoder approach was benchmarked against standard statistical imputation tools.

Main Results:

• The autoencoder-based imputation achieved superior accuracy across various allele frequencies and diverse ancestries.
• Inference run time was at least fourfold faster compared to standard imputation methods.

Conclusions:

• Autoencoder-based genotype imputation offers a more accurate, faster, and accessible alternative to traditional methods.
• This approach has the potential to democratize advanced imputation techniques in population genetics and genomics studies.