Fast spatial ancestry via flexible allele frequency surfaces.

Area of Science:

• Population Genetics
• Computational Biology
• Bioinformatics

Background:

• Determining geographic origins from genetic data presents unique computational and modeling challenges.
• Single-nucleotide polymorphisms (SNPs) have variable informativeness, and allele frequencies shift non-linearly with geography.
• Existing methods struggle with integrating multi-SNP evidence and accurately localizing individuals of mixed ancestry.

Purpose of the Study:

• To develop a novel computational model for estimating the geographic origins of individuals using genetic data.
• To address challenges in SNP informativeness, non-linear allele frequency changes, and the integration of multi-SNP evidence.
• To accurately localize individuals with both unmixed and admixed ancestry.

Main Methods:

• A novel model inspired by image processing and optimization theory is proposed, dividing regions into pixels.
• Allele frequencies are estimated across landscapes by maximizing a penalized product of binomial likelihoods using a minorize-maximize (MM) algorithm.
• Bayes' rule is applied to compute posterior probabilities for origin pixels, with a penalized MM algorithm handling admixed individuals.

Main Results:

• The model was applied to the Population Reference Sample (POPRES) data, demonstrating superior localization accuracy for both unmixed and admixed individuals.
• The method achieves better localization than existing approaches while utilizing only a small fraction of available SNPs.
• Computational performance is comparable to the best existing competing software.

Conclusions:

• The developed model offers a significant advancement in estimating geographic origins from genetic data.
• The approach is robust for both homogeneous and admixed populations.
• The OriGen R package will be made freely available, facilitating broader adoption and research.