Mind the Gap: A Neural Network Framework for Imputing Genotypes in Non-Model Species.

Area of Science:

• Genomics
• Bioinformatics
• Population Genetics

Background:

• Reduced representation sequencing (RRS) is cost-effective for large-scale studies in non-model species.
• RRS often results in substantial missing genotype data, reducing statistical power and biasing analyses.
• Genotype imputation can overcome missing data issues but typically requires a reference panel, which is often unavailable for non-model species.

Purpose of the Study:

• To develop and evaluate gtImputation, an unsupervised machine learning tool for genotype imputation.
• To enable accurate imputation without the need for a reference panel in non-model species.
• To improve the utility of RRS data for downstream population genetic analyses.

Main Methods:

• Development of gtImputation, an unsupervised machine learning algorithm with a graphical user interface (GUI).
• The tool leverages the inherent data structure for imputation, bypassing the need for external reference panels.
• Performance evaluation against existing imputation methods, including haplotype-clustering and other unsupervised machine learning algorithms.

Main Results:

• gtImputation demonstrates comparable or superior performance to existing methods.
• The tool is particularly effective for imputing sites with low minor allele frequency (MAF).
• Performance is enhanced in datasets exhibiting strong population structure.

Conclusions:

• gtImputation provides a robust, reference-panel-free solution for genotype imputation in non-model species.
• This method expands the applicability of imputation, enabling complex analyses with RRS data.
• The tool facilitates lower sequencing costs while maintaining high data quality for genomic studies.