Local Ancestry Inference in Large Pedigrees

Heming Wang,Xiang Zhang,Xiaofeng Zhu,Robert C Elston,Susan Redline,Tamar Sofer

doi:10.1038/s41598-019-57039-w

Abstract

Local ancestry, defined as the genetic ancestry at a genomic location of an admixed individual, is widely used as a genetic marker in genetic association and evolutionary genetics studies. Many methods have been developed to infer the local ancestries in a set of unrelated individuals, a few of them have been extended to small nuclear families, but none can be applied to large (e.g. three-generation) pedigrees. In this study, we developed a method, FamANC, that can improve the accuracy of local ancestry inference in large pedigrees by: (1) using an existing algorithm to infer local ancestries for all individuals in a family, assuming (contrary to fact) they are unrelated, and (2) improving its accuracy by correcting inference errors using pedigree structure. Applied on African-American pedigrees from the Cleveland Family Study, FamANC was able to correct all identified Mendelian errors and most of double crossovers.

Highlights

There has been an increasing interest in studying the ancestral spectrum of admixed individuals, such as African Americans and Latino Americans1–3
All the above methods can be applied to phased haplotypes or diploid genotypes using a joint Hidden Markov Model (HMM) applied to the two haplotypes in an admixed individual
We modeled the probabilities of observing different numbers of individuals with ancestry inference errors at one locus in a family using a binomial distribution with inference error probability 0.01 (Supplementary Fig. 1)

Summary

Introduction

There has been an increasing interest in studying the ancestral spectrum of admixed individuals, such as African Americans and Latino Americans. Hidden Markov Model (HMM)-based methods, including SABER+13, HAPMIX14, LAMP-LD15, and PCAdmix, are widely used because of their high accuracy and resolution These methods model the observed genotypes of admixed individuals conditioning on the hidden states of their ancestral reference alleles or haplotypes, which are assumed to follow a Markov process. The common approach for inferring local ancestries in large pedigrees is currently to incorrectly assume individuals are unrelated This approach may result in multiple Mendelian errors that violate the assumption of family-based genetic analyses. We developed a method which estimates local ancestries in large pedigrees by: [1] using existing software (e.g. SABER+ and HAPMIX) to infer local ancestries for all individuals in a family, temporarily assuming they are unrelated, and [2] using FamANC to improve the local ancestry inference accuracy by using the known pedigree structure to correct inference errors

Methods

Results

Conclusion