Benchmarking kinship estimation tools for ancient genomes using pedigree simulations.

Area of Science:

• Ancient Genomics
• Population Genetics
• Bioinformatics

Background:

• Growing interest in reconstructing ancient kinship patterns using low-coverage palaeogenomes.
• Need for robust kinship estimation tools applicable to limited ancient genetic data.

Purpose of the Study:

• To benchmark and compare the performance of four kinship estimation tools (lcMLkin, NgsRelate, KIN, READ) on low-coverage palaeogenomic data.
• To evaluate tool accuracy with varying numbers of single nucleotide polymorphisms (SNPs) and assess the impact of population allele frequency noise and inbreeding.

Main Methods:

• Utilized pedigree and ancient genome sequence simulations with limited shared SNPs (1 to 50 K, MAF ≥0.01).
• Assessed kinship estimation accuracy for different degrees of relatedness (first, second, third-degree relatives).
• Investigated the influence of population allele frequency noise and inbreeding on kinship coefficient estimations.

Main Results:

• All four tools performed comparably with ≥20 K SNPs.
• First-degree relatives were accurately classified with as few as 1 K SNPs (F1 scores up to 96% with NgsRelate/lcMLkin).
• Distinguishing third-degree relatives was highly accurate (F1 >90%) with 5 K SNPs using NgsRelate and lcMLkin, outperforming READ and KIN.

Conclusions:

• NgsRelate and lcMLkin demonstrate high accuracy for close and distant relatives using limited SNPs in palaeogenomic data.
• Population allele frequency noise and inbreeding can affect kinship coefficients, with varying tool sensitivities.
• Employing multiple kinship estimation tools in parallel is recommended for robust results with ultra-low-coverage ancient genomes.