Localizing Post-Admixture Adaptive Variants with Object Detection on Ancestry-Painted Chromosomes

Area of Science:

• Population genetics
• Genomics
• Bioinformatics

Background:

• Gene flow during population admixture can introduce adaptive alleles.
• Genetic ancestry patterns are used to detect positive selection after admixture.
• Current methods for detecting selection based on local ancestry have limitations, including potential false positives/negatives and broad genomic inferences.

Purpose of the Study:

• To develop a new computational method for identifying local ancestry outliers indicative of positive selection.
• To overcome limitations of existing methods by incorporating genomic context and avoiding user-defined statistics.
• To provide a more precise and biologically interpretable way to detect selection in admixed populations.

Main Methods:

• Developed a deep learning object detection model applied to local ancestry-painted genome images.
• Utilized simulated data to test the robustness of the method under various demographic scenarios.
• Applied the method to human genotype data from Cabo Verde.

Main Results:

• The deep learning method demonstrated robustness to demographic misspecifications in simulations.
• The approach successfully localized a known adaptive locus in Cabo Verdean human data to a narrow genomic region.
• Compared to existing ancestry-based methods, this new technique provided more precise localization of selection signals.

Conclusions:

• The developed deep learning method offers a more accurate and refined approach to detecting positive selection in admixed populations.
• This method improves upon traditional techniques by leveraging genomic context and reducing the inference of wide genomic regions under selection.
• The findings have implications for understanding adaptation and genetic variation in admixed populations, enhancing biological interpretation.