BREADR: An R Package for the Bayesian Estimation of Genetic Relatedness from Low-coverage Genotype Data

Abstract

Robust and reliable estimates of how individuals are biologically related to each other are a key source of information when reconstructing pedigrees. In combination with contextual data, reconstructed pedigrees can be used to infer possible kinship practices in prehistoric populations. However, standard methods to estimate biological relatedness from genome sequence data cannot be applied to low coverage sequence data, such as are common in ancient DNA (aDNA) studies. Critically, a statistically robust method for assessing and quantifying the confidence of a classification of a specific degree of relatedness for a pair of individuals, using low coverage genome data, is lacking. In this paper we present the R-package BREADR (Biological RElatedness from Ancient DNA in R), which leverages the so-called pairwise mismatch rate, calculated on optimally-thinned genome-wide pseudo-haploid sequence data, to estimate genetic relatedness up to the second degree, assuming an underlying binomial distribution. BREADR also returns a posterior probability for each degree of relatedness, from identical twins/same individual, first-degree, second-degree or "unrelated" pairs, allowing researchers to quantify and report the uncertainty, even for particularly low-coverage data. We show that this method accurately recovers degrees of relatedness for sequence data with coverage as low as 0.04X using simulated data, and then compare the performance of BREADR on empirical data from Bronze Age Iberian human sequence data. The BREADR package is designed for pseudo-haploid genotype data, common in aDNA studies.

READv2: Advanced and user-friendly detection of biological relatedness in archaeogenomics

“…We introduced READv2 as a version with increased efficiency and compared it to READv1. Other studies have already covered comparing the general READ approach to other methods used in the field [31][32][33]43,38]. Methods such as ancIBD [35], KIN [32], TKGWV2 [31] as well as the genotype likelihood-based lcMLkin [29] and ngsRelate [30] have specific advantages, either providing more precise results with lower amounts of data or by being able to detect higher than second-degree relatedness confidently.…”

“…ancIBD [35] and KIN [32] were both specifically designed for ancient DNA data and their HMM approaches allow for the classification of higher degrees of relatedness as well as the differentiation between siblings and parent-offspring pairs. READv2 is very similar in its approach to BREADR [33] and TKGWV2 [31], with each tool having its own unique feature. READv2 has the functionality to separate the different first-degree relationships, BREADR has a better quantification of uncertainty, and TKGWV2 works well with lower amounts of input data.…”

“…The authors first ran READ [27] to estimate the degree of relatedness for each pair of individuals, followed by lcMLkin [29] to differentiate between parent-offspring and siblings among first-degree relationships. They further used BREADR [33] for individual pairs as well as imputation and ancIBD [35] for higher degrees. We have already shown that READv2 can process this dataset a lot faster than READv1 (Figure 1A).…”

“…Recent years have seen the introduction or application of more advanced methods into the field which work with lower amounts of data (Table S1), provide resolution for more distant degrees of relatedness, and/or are able to differentiate between different types of relationships for the same degree (e.g. parent-offspring versus full siblings) [29][30][31][32][33][34][35].…”

Evidence for dynastic succession among early Celtic elites in Central Europe

“…We calculated the pairwise mismatch rate 60 in all pairs of individuals from our pseudo-haploid dataset to double-check for potential duplicate individuals and to determine first-, secondand third-degree relatives. For this purpose, we also used BREADR 101 which utilizes Bayesian posterior probabilities for the classification of the genetic relationships. Additionally, we also applied LcMLkin 102 (v0.5.0) and KIN 103 (v3.1.3), which use genotype likelihoods to estimate the three k coefficients (k 0 , k 1 or k 2 ), which define the probability that two individuals have zero, one or two alleles identical by descent at a random site in the genome (Supplementary Note 2).…”