Geography and genography: prediction of continental origin using randomly selected single nucleotide polymorphisms
- Dominic J Allocco1, 2Email author,
- Qing Song3,
- Gary H Gibbons3,
- Marco F Ramoni†1, 4 and
- Isaac S Kohane†1, 4
© Allocco et al; licensee BioMed Central Ltd. 2007
Received: 29 June 2006
Accepted: 10 March 2007
Published: 10 March 2007
Recent studies have shown that when individuals are grouped on the basis of genetic similarity, group membership corresponds closely to continental origin. There has been considerable debate about the implications of these findings in the context of larger debates about race and the extent of genetic variation between groups. Some have argued that clustering according to continental origin demonstrates the existence of significant genetic differences between groups and that these differences may have important implications for differences in health and disease. Others argue that clustering according to continental origin requires the use of large amounts of genetic data or specifically chosen markers and is indicative only of very subtle genetic differences that are unlikely to have biomedical significance.
We used small numbers of randomly selected single nucleotide polymorphisms (SNPs) from the International HapMap Project to train naïve Bayes classifiers for prediction of ancestral continent of origin. Predictive accuracy was tested on two independent data sets. Genetically similar groups should be difficult to distinguish, especially if only a small number of genetic markers are used. The genetic differences between continentally defined groups are sufficiently large that one can accurately predict ancestral continent of origin using only a minute, randomly selected fraction of the genetic variation present in the human genome. Genotype data from only 50 random SNPs was sufficient to predict ancestral continent of origin in our primary test data set with an average accuracy of 95%. Genetic variations informative about ancestry were common and widely distributed throughout the genome.
Accurate characterization of ancestry is possible using small numbers of randomly selected SNPs. The results presented here show how investigators conducting genetic association studies can use small numbers of arbitrarily chosen SNPs to identify stratification in study subjects and avoid false positive genotype-phenotype associations. Our findings also demonstrate the extent of variation between continentally defined groups and argue strongly against the contention that genetic differences between groups are too small to have biomedical significance.
A major goal of both the Human Genome Project and the subsequent International HapMap Project was to provide a foundation for understanding genetic variation in the human genome [1, 2]. Multiple studies over the past several decades have consistently concluded that only 5–15% of human genetic variation can be explained by differences between populations [3–6]. Although this proportion is relatively small, recent studies have shown that when individuals are grouped on the basis of genetic similarity, group membership corresponds closely to ancestral geographic origin  or self-identified race or ethnicity .
There has been considerable discussion about the implications of these findings in the context of larger debates about population sub-structure, race and genetics [8–13]. One major aspect of the debate concerns the extent of genetic variation between groups. This debate is becoming less theoretical as targeted therapies like BiDil (a branded combination of hydralazine and isosorbide dinitrate approved by the FDA for the treatment of congestive heart failure in African-Americans) are introduced. Some have argued that the genetic differences between continentally defined groups are relatively small and thus unlikely to have biomedical significance [3, 14]. Providing support for this view, others have noted that accurate classification of ancestral origin, while possible, requires large numbers of genetic markers [9–12, 15]. The opposing viewpoint holds that there are significant genetic differences between groups and that these genetic differences may account for differences in risk of disease between populations . Proponents of this viewpoint also note that an understanding of genetic variation between groups is important to avoid confounding in genotype-phenotype association studies [7, 17].
A variety of different computational methods, including those implemented in the popular programs STRUCTURE and BAPS, have been developed and used to explore genetic variation in populations [18–20]. Previous investigators have used variation in both mitochondrial DNA and microsatellites to characterize population sub-structure and ancestral geographic origin [5, 7, 21–23]. Analysis of the relationship between genetic variation in chromosomal (as opposed to mitochondrial) SNPs and ancestral geographic origin has been more limited because large scale, genome-wide SNP data from geographically diverse individuals has not been available. Lao et al recently used data from Affymetrix 10K arrays to identify 10 specific SNPs that were highly informative for characterizing ancestry. These SNPs were found to be somewhat less informative when used to characterize an independent data set.
In this paper, we use recently published genome-wide SNP data to analyze population sub-structure and ancestral geographic origin. The analysis uses a much larger number of SNPs than has previously been possible and focuses on the use of randomly selected SNPs. We show that one can accurately predict continent of origin in independent data sets using only a very small number of randomly selected SNPs. Genotype data from 50 random SNPs is sufficient to predict ancestral continent of origin in our primary test data set with an average accuracy of 95%. We analyze the comparative utility of SNPs in introns, coding exons, regulatory regions and regions coding for untranslated mRNA for prediction of ancestry. Finally, we show that SNPs that are informative about ancestry are common and widely distributed throughout the genome. Our findings demonstrate how researchers conducting SNP based genotype-phenotype association studies can accurately and reproducibly characterize ancestry using random SNPs, as opposed to SNPs specifically chosen to be informative about ancestry. As genetically similar groups should be difficult to distinguish using only a small number of genetic markers, our results also demonstrate the extent of genetic variation between continentally defined groups and argue strongly against the contention that differences between groups are too small to have biomedical significance.
The International HapMap Project is a large collaborative effort that has made publicly available genotype data for 270 individuals from four different populations: Yoruba in Ibadan, Nigeria (YRI); Japanese in Tokyo, Japan (JPT); Han Chinese in Beijing (CHB), China; and Utah residents with ancestry from northern and western Europe (CEU) . At the time of our study, genotype data that had passed quality control filters was available for almost 4 million SNPs. We used the HapMap data to develop classifiers for predicting ancestral continent of origin and tested these classifiers on independent data sets.
Our primary test data set consisted of 1,586,383 SNPs genotyped by investigators at Perlegen Sciences to study DNA variation in human populations . We excluded nine individuals of European ancestry in this data because they were also genotyped in the HapMap Project. Thus, this test data set included genotype data for 23 African-Americans, 15 European-Americans and 24 Han Chinese. The second test data set consisted of 4,124 SNPs genotyped as part of the Innate Immunity Program for Genomic Applications (IIPGA) and made publicly available on their website . Nine individuals in this data set were also excluded from the analysis because they were genotyped in the HapMap Project. This test data set therefore included data for 24 African-Americans and 14 European-Americans. SNPs in the HapMap and Perlegen data sets were selected so as to be generally representative of variation throughout the genome, while SNPs in the IIPGA data set were selected by the original investigators for genotyping on the basis of potential involvement in the innate immune response.
For each of the test data sets, we limited our analysis to the bi-allelic SNPs that were present on autosomal chromosomes in both the HapMap data and the test set data. There were 1,047,543 and 1,588 such SNPs in the intersection of the HapMap data with the Perlegen and IIPGA data, respectively.
In this paper, we use naïve Bayes classifiers trained on data from the HapMap to predict the ancestral geographic origin of individuals from three independent data sets. Even when a relatively small number of randomly selected SNPs are used, classification is accurate and robust. In the large Perlegen data set, predictive accuracy increases to 100% as the number of SNPs grows. This is not the case for the IIPGA test data set. Predictive accuracy as a function of the number of SNPs used, while still very good in comparison to many tests used in biology and medicine, levels off at 95% with two individuals being consistently misclassified. These two individuals were classified incorrectly even if all 1588 available SNPs were used to train the classifier. The two misclassifications were the two African-Americans labeled with arrows in Fig. 2 who were noted to be intermediate between the African and European clusters. These two individuals may have mixed African and European ancestry or may have African ancestors from different parts of Africa than the Yoruba used to train the classifier.
Accurate characterization of ancestry will allow investigators conducting genetic association studies to identify stratification in study subjects and avoid false positive genotype-phenotype associations. The analysis we describe here is designed to predict ancestral continent of origin. It could easily be extended to make predictions about smaller units of geography or individuals with a mixed background. This would require more extensive genotype data and well-characterized information about ancestral geographic origin from such individuals. There is only very limited data of this kind currently available, but this is expected to change in the future as genotyping costs decrease. Thus we anticipate that identification of more complex patterns of ancestry will be increasingly feasible as the amount of available data grows. This in turn will allow the development of higher resolution genographic maps and provide investigators designing genetic association studies with more powerful tools for detecting stratification.
In this paper, we have shown that the differences between continentally defined groups are sufficiently large that even a randomly selected, minute fraction of the genetic variation in the human genome can be used to characterize ancestral geographic origin in an accurate and reproducible manner. This argues strongly against the contention that differences between groups are too small to have biomedical significance. Understanding if and how these differences relate to risk of disease and response to therapy is one of the major challenges facing the biomedical research community.
Some have argued that the differences between continentally defined groups are relatively small and that it is difficult to distinguish groups without using large amounts of genetic data or specifically chosen markers. Our results show that continentally defined groups can be easily distinguished using only a small number of randomly selected SNPs. SNPs that are informative about ancestry are common and widely distributed throughout the genome and across SNP types. These findings illustrate the extent of genetic variation between continentally defined groups.
The HapMap, Perlegen and IIPGA data sets were obtained from their respective websites [25, 30, 31]. For the HapMap data we used the non-redundant data sets from public release 19 which contained data for phases I and II. Thirty trios were genotyped for both the Yoruba and European populations. We excluded the children from our analysis because they did not represent independent genotypes. We also excluded the one individual from the Japanese population who did not have phase I data. There were 60 Yoruba, 60 European-Americans, 45 Han Chinese and 44 Japanese. For the purpose of our analysis, the Han Chinese and Japanese subjects were grouped together.
For the IIPGA data set, we used dbSNP annotation files to map the IIPGA identifiers to the official NCBI reference SNP identifier . The dbSNP annotation files were also used to determine SNP type.
Nine individuals in both the Perlegen and IIPGA data sets were also genotyped as part of the HapMap project. In our study, these individuals were included in the HapMap data set, but not in the Perlegen or IIPGA data sets.
The genetic distance between two individuals at a single loci was defined to be zero if the two individuals had the same genotype, 1/2 if they had one allele in common and 1 if they had neither allele in common (i.e. d(CG,CC) = .5 and d(CC,GG) = 1) . The genetic distance between two individuals was calculated as the mean genetic distance over all loci genotyped in both individuals. The genetic distance between an individual and a group was defined as the mean of the pairwise genetic distances between the individual and all members of the group – except if the individual was a member of the group, we did not include the distance between the individual and him or herself. Genetic distances were normalized so that for each individual the genetic distances to the HapMap YRI, CEU and JPT+CHB summed to one.
Classification with naïve Bayes
We wrote our own implementation of the naïve Bayes algorithm  in PERL. We assumed a uniform prior probability distribution over the class variable. For each test data set, we randomly selected a specified number of SNPs. We then used the HapMap data for these SNPs to train a classifier. The performance of this classifier was then evaluated by determining accuracy of prediction on the test data set. This process was repeated 100 times for the specified number of SNPs.
Identification of SNPs informative about ancestry
To evaluate the distribution of SNPs where allele frequency differed significantly between groups, we computed the pairwise differences in major allele frequency among the three HapMap groups. For the purposes of our study, the most frequent allele in the Yoruba was considered to be the major allele. For each pairwise combination of groups, we determined the proportion of SNPs where the difference in major allele frequency exceeded a threshold value. We investigated how this proportion varied across the genome.
We also used the informativeness for assignment measure to analyze SNPs . The informativeness for assignment (I) of a SNP is defined as:
where there are N alleles and K populations, p ij represents the frequency of allele j in population i, and p j represents the mean value of p ij over the K populations. This measure is a generalization of the absolute difference in allele frequency to more than two populations.
We analyzed the distribution of informative SNPs on both a chromosomal basis and by dividing the chromosomes into bins of a constant size. We tested multiple combinations of bin size and allele frequency cutoff threshold. The conclusion that SNPs useful for classification are distributed across the genome was not sensitive to changes in these parameters.
One-way ANOVA (calculated using Microsoft Excel) was used to compare mean predictive accuracy across SNP types and to compare informativeness for assignment across both chromosomes and bins. The null hypothesis was that all means were equal. P-values less than .05 were considered to be statistically significant.
This work was supported by grants from the National Science Foundation, the National Library of Medicine and the National Heart, Lung and Blood Institute of the National Institutes of Health.
- Collins FS, Patrinos A, Jordan E, Chakravarti A, Gesteland R, Walters L: New goals for the U.S. Human Genome Project: 1998-2003. Science. 1998, 282 (5389): 682-689. 10.1126/science.282.5389.682.PubMedView ArticleGoogle Scholar
- The International HapMap Project. Nature. 2003, 426 (6968): 789-796. 10.1038/nature02168.Google Scholar
- Lewontin RC: The apportionment of human diversity. Evolutionary Biology. Edited by: Dobzhansky TH, Hecht MK, Steere WC. 1972, Appelton-Century-Crofts, 6: 381-398.View ArticleGoogle Scholar
- Nei M, Roychoudhury AK: Genic variation within and between the three major races of man, Caucasoids, Negroids, and Mongoloids. Am J Hum Genet. 1974, 26 (4): 421-443.PubMed CentralPubMedGoogle Scholar
- Rosenberg NA, Pritchard JK, Weber JL, Cann HM, Kidd KK, Zhivotovsky LA, Feldman MW: Genetic structure of human populations. Science. 2002, 298 (5602): 2381-2385. 10.1126/science.1078311.PubMedView ArticleGoogle Scholar
- Cavalli-Sforza LL, Feldman MW: The application of molecular genetic approaches to the study of human evolution. Nat Genet. 2003, 33 Suppl: 266-275. 10.1038/ng1113.PubMedView ArticleGoogle Scholar
- Tang H, Quertermous T, Rodriguez B, Kardia SL, Zhu X, Brown A, Pankow JS, Province MA, Hunt SC, Boerwinkle E, Schork NJ, Risch NJ: Genetic structure, self-identified race/ethnicity, and confounding in case-control association studies. Am J Hum Genet. 2005, 76 (2): 268-275. 10.1086/427888.PubMed CentralPubMedView ArticleGoogle Scholar
- Haga SB, Venter JC: Genetics. FDA races in wrong direction. Science. 2003, 301 (5632): 466-10.1126/science.1087004.PubMedView ArticleGoogle Scholar
- Calafell F: Classifying humans. Nat Genet. 2003, 33 (4): 435-436. 10.1038/ng0403-435.PubMedView ArticleGoogle Scholar
- Tishkoff SA, Kidd KK: Implications of biogeography of human populations for 'race' and medicine. Nat Genet. 2004, 36 (11 Suppl): S21-7. 10.1038/ng1438.PubMedView ArticleGoogle Scholar
- Jorde LB, Wooding SP: Genetic variation, classification and 'race'. Nat Genet. 2004, 36 (11 Suppl): S28-33. 10.1038/ng1435.PubMedView ArticleGoogle Scholar
- Serre D, Paabo S: Evidence for gradients of human genetic diversity within and among continents. Genome Res. 2004, 14 (9): 1679-1685. 10.1101/gr.2529604.PubMed CentralPubMedView ArticleGoogle Scholar
- Risch N: Dissecting racial and ethnic differences. N Engl J Med. 2006, 354 (4): 408-411. 10.1056/NEJMe058265.PubMedView ArticleGoogle Scholar
- Cooper RS, Kaufman JS, Ward R: Race and genomics. N Engl J Med. 2003, 348 (12): 1166-1170. 10.1056/NEJMsb022863.PubMedView ArticleGoogle Scholar
- Lao O, van Duijn K, Kersbergen P, de Knijff P, Kayser M: Proportioning whole-genome single-nucleotide-polymorphism diversity for the identification of geographic population structure and genetic ancestry. Am J Hum Genet. 2006, 78 (4): 680-690. 10.1086/501531.PubMed CentralPubMedView ArticleGoogle Scholar
- Burchard EG, Ziv E, Coyle N, Gomez SL, Tang H, Karter AJ, Mountain JL, Perez-Stable EJ, Sheppard D, Risch N: The importance of race and ethnic background in biomedical research and clinical practice. N Engl J Med. 2003, 348 (12): 1170-1175. 10.1056/NEJMsb025007.PubMedView ArticleGoogle Scholar
- Risch N, Burchard E, Ziv E, Tang H: Categorization of humans in biomedical research: genes, race and disease. Genome Biol. 2002, 3 (7): 1-12. 10.1186/gb-2002-3-7-comment2007.View ArticleGoogle Scholar
- Corander J, Waldmann P, Sillanpaa MJ: Bayesian analysis of genetic differentiation between populations. Genetics. 2003, 163 (1): 367-374.PubMed CentralPubMedGoogle Scholar
- Dawson KJ, Belkhir K: A Bayesian approach to the identification of panmictic populations and the assignment of individuals. Genet Res. 2001, 78 (1): 59-77. 10.1017/S001667230100502X.PubMedView ArticleGoogle Scholar
- Pritchard JK, Stephens M, Donnelly P: Inference of population structure using multilocus genotype data. Genetics. 2000, 155 (2): 945-959.PubMed CentralPubMedGoogle Scholar
- Parra EJ, Marcini A, Akey J, Martinson J, Batzer MA, Cooper R, Forrester T, Allison DB, Deka R, Ferrell RE, Shriver MD: Estimating African American admixture proportions by use of population-specific alleles. Am J Hum Genet. 1998, 63 (6): 1839-1851. 10.1086/302148.PubMed CentralPubMedView ArticleGoogle Scholar
- Salas A, Richards M, Lareu MV, Scozzari R, Coppa A, Torroni A, Macaulay V, Carracedo A: The African diaspora: mitochondrial DNA and the Atlantic slave trade. Am J Hum Genet. 2004, 74 (3): 454-465. 10.1086/382194.PubMed CentralPubMedView ArticleGoogle Scholar
- Wilson JF, Weale ME, Smith AC, Gratrix F, Fletcher B, Thomas MG, Bradman N, Goldstein DB: Population genetic structure of variable drug response. Nat Genet. 2001, 29 (3): 265-269. 10.1038/ng761.PubMedView ArticleGoogle Scholar
- Hinds DA, Stuve LL, Nilsen GB, Halperin E, Eskin E, Ballinger DG, Frazer KA, Cox DR: Whole-genome patterns of common DNA variation in three human populations. Science. 2005, 307 (5712): 1072-1079. 10.1126/science.1105436.PubMedView ArticleGoogle Scholar
- Innate Immunity PGA, NHLBI Program for Genomic Applications. [http://innateimmunity.net]
- Mountain JL, Cavalli-Sforza LL: Multilocus genotypes, a tree of individuals, and human evolutionary history. Am J Hum Genet. 1997, 61 (3): 705-718.PubMed CentralPubMedView ArticleGoogle Scholar
- Domingos P, Pazzani M: On the optimality of the simple Bayesian classifier under zero-one loss. Machine Learning. 1997, 29: 103-130. 10.1023/A:1007413511361.View ArticleGoogle Scholar
- Smigielski EM, Sirotkin K, Ward M, Sherry ST: dbSNP: a database of single nucleotide polymorphisms. Nucleic Acids Res. 2000, 28 (1): 352-355. 10.1093/nar/28.1.352.PubMed CentralPubMedView ArticleGoogle Scholar
- Rosenberg NA, Li LM, Ward R, Pritchard JK: Informativeness of genetic markers for inference of ancestry. Am J Hum Genet. 2003, 73 (6): 1402-1422. 10.1086/380416.PubMed CentralPubMedView ArticleGoogle Scholar
- The International HapMap Project. [http://www.hapmap.org]
- Perlegen Sciences Genotype Browser. [http://genome.perlegen.com]
- Maron M: Automatic Indexing: An Experimental Inquiry. JACM. 1961, 8 (3): 404-417. 10.1145/321075.321084.View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.