Prediction of biogeographical ancestry from genotype: a comparison of classifiers.

DNA can provide forensic intelligence regarding a donor's biogeographical ancestry (BGA) and other externally visible characteristics (EVCs). A number of algorithms have been proposed to assign individual human genotypes to a BGA using ancestry informative marker (AIM) panels. This study compares the BGA assignment accuracy of the population clustering program STRUCTURE and three generic classification approaches including a Bayesian algorithm, genetic distance, and multinomial logistic regression (MLR). A selection of 142 ancestry informative single nucleotide polymorphisms (SNPs) were chosen from existing marker panels (SNPforID 34-plex, Eurasiaplex, Seldin, and Kidd's AIM panels) to assess BGA classification at the continental level for Africans, Europeans, East Asians, and Amerindians. A training set of 1093 individuals with self-declared BGA from the 1000 Genomes phase 1 database was used by each classifier to predict BGA in a test set of 516 individuals from the HGDP-CEPH (Stanford) cell line panel. Tests were repeated with 0, 10, 50, 70, and 90% of the genotypes missing. Comparison of the area under the receiver operating characteristic curves (AUROCs) showed high accuracy in STRUCTURE and the generic Bayesian approach. The latter algorithm offers a computationally simpler alternative to STRUCTURE with little loss in accuracy and is suitable for phenotype prediction while STRUCTURE is not.

Development and Evaluation of the Ancestry Informative Marker Panel of the VISAGE Basic Tool.

We detail the development of the ancestry informative single nucleotide polymorphisms (SNPs) panel forming part of the VISAGE Basic Tool (BT), which combines 41 appearance predictive SNPs and 112 ancestry predictive SNPs (three SNPs shared between sets) in one massively parallel sequencing (MPS) multiplex, whereas blood-based age analysis using methylation markers is run in a parallel MPS analysis pipeline. The selection of SNPs for the BT ancestry panel focused on established forensic markers that already have a proven track record of good sequencing performance in MPS, and the overall SNP multiplex scale closely matched that of existing forensic MPS assays. SNPs were chosen to differentiate individuals from the five main continental population groups of Africa, Europe, East Asia, America, and Oceania, extended to include differentiation of individuals from South Asia. From analysis of 1000 Genomes and HGDP-CEPH samples from these six population groups, the BT ancestry panel was shown to have no classification error using the Bayes likelihood calculators of the Snipper online analysis portal. The differentiation power of the component ancestry SNPs of BT was balanced as far as possible to avoid bias in the estimation of co-ancestry proportions in individuals with admixed backgrounds. The balancing process led to very similar cumulative population-specific divergence values for Africa, Europe, America, and Oceania, with East Asia being slightly below average, and South Asia an outlier from the other groups. Comparisons were made of the African, European, and Native American estimated co-ancestry proportions in the six admixed 1000 Genomes populations, using the BT ancestry panel SNPs and 572,000 Affymetrix Human Origins array SNPs. Very similar co-ancestry proportions were observed down to a minimum value of 10%, below which, low-level co-ancestry was not always reliably detected by BT SNPs. The Snipper analysis portal provides a comprehensive population dataset for the BT ancestry panel SNPs, comprising a 520-sample standardised reference dataset; 3445 additional samples from 1000 Genomes, HGDP-CEPH, Simons Foundation and Estonian Biocentre genome diversity projects; and 167 samples of six populations from in-house genotyping of individuals from Middle East, North and East African regions complementing those of the sampling regimes of the other diversity projects.

Performance of ancestry-informative SNP and microhaplotype markers.

The use of microhaplotypes (MHs) for ancestry inference has added to an increasing number of ancestry-informative markers (AIMs) for forensic application that includes autosomal single nucleotide polymorphisms (SNPs) and insertions/deletions (indels). This study compares bi-allelic and tri-allelic SNPs as well as MH markers for their ability to differentiate African, European, South Asian, East Asian, and American population groups from the 1000 Genomes Phase 3 database. A range of well-established metrics were applied to rank each marker according to the population differentiation potential they measured. These comprised: absolute allele frequency differences (δ); Rosenberg's informativeness for (ancestry) assignment (In); the fixation index (FST); and the effective number of alleles (Ae). A panel consisting of all three marker types resulted in the lowest mean divergence per population per individual (MDPI = 2.16%) when selected by In. However, when marker types were not mixed, MHs were the highest performing markers by most metrics (MDPI < 4%) for differentiation between the five continental populations.

Prediction of biogeographical ancestry in admixed individuals.

Estimation of ancestral affiliation for human genotypes is now possible for major geographic populations and has been employed for forensic casework. Prediction algorithms, such as the Snipper Bayesian classifier, have the ability to classify non-admixed BGA in African (AFR), European (EUR), East Asian (EAS), and most Amerindian (NAM) individuals, but are not always appropriate for admixed individuals. Artificial admixture was simulated for all possible admixture ratios (1:1, 3:1, 2:1:1, and 1:1:1:1) from four grandparents. The simulated genotypes were used to test the accuracy of various prediction algorithms, most successful of which were the population genetics program, STRUCTURE, and a novel genetic distance algorithm (GDA). STRUCTURE was ideal for admixed individuals with 1:1 and 3:1 ratios from AFR, EUR, EAS, and NAM reference populations. Individuals with 1:1:1:1 BGA proportions were more accurately predicted by GDA. The use of hypothetical root genotypes improved the accuracy of GDA predictions for 1:1 and 3:1 admixtures and STRUCTURE classification of 1:1:1:1 admixture. The GDA requires only allele or genotype frequency values from each reference population, which offers a simpler sampling and input formatting procedure than is required by STRUCTURE. It can also be implemented in a spreadsheet without the need for long run times.