BMC Genetics

, 14:38 | Cite as

Genotype imputation accuracy in a F2 pig population using high density and low density SNP panels

  • Jose L Gualdrón Duarte
  • Ronald O Bates
  • Catherine W Ernst
  • Nancy E Raney
  • Rodolfo JC Cantet
  • Juan P Steibel
Open Access
Research article
Part of the following topical collections:
  1. Statistical and computational genetics



F2 resource populations have been used extensively to map QTL segregating between pig breeds. A limitation associated with the use of these resource populations for fine mapping of QTL is the reduced number of founding individuals and recombinations of founding haplotypes occurring in the population. These limitations, however, become advantageous when attempting to impute unobserved genotypes using within family segregation information. A trade-off would be to re-type F2 populations using high density SNP panels for founding individuals and low density panels (tagSNP) in F2 individuals followed by imputation. Subsequently a combined meta-analysis of several populations would provide adequate power and resolution for QTL mapping, and could be achieved at relatively low cost. Such a strategy allows the wealth of phenotypic information that has previously been obtained on experimental resource populations to be further mined for QTL identification. In this study we used experimental and simulated high density genotypes (HD-60K) from an F2 cross to estimate imputation accuracy under several genotyping scenarios.


Selection of tagSNP using physical distance or linkage disequilibrium information produced similar imputation accuracies. In particular, tagSNP sets averaging 1 SNP every 2.1 Mb (1,200 SNP genome-wide) yielded imputation accuracies (IA) close to 0.97. If instead of using custom panels, the commercially available 9K chip is used in the F2, IA reaches 0.99. In order to attain such high imputation accuracy the F0 and F1 generations should be genotyped at high density. Alternatively, when only the F0 is genotyped at HD, while F1 and F2 are genotyped with a 9K panel, IA drops to 0.90.


Combining 60K and 9K panels with imputation in F2 populations is an appealing strategy to re-genotype existing populations at a fraction of the cost.


Linkage Disequilibrium Minor Allele Frequency Imputation Accuracy Impute Genotype High Density Genotype 


The search for regions in the genome containing genetic variants that affect production traits requires experimental populations to identify the segregating QTL within and between parental populations [1]. The F2 design is commonly used to map QTL segregating in divergent parental lines [2, 3]. To produce reliable analyses of association or genetic evaluations using genomic information, a great number of individuals with phenotypes and high density (HD) genotypes are required [4]. However, HD genotypes for large numbers of animals are expensive to obtain [5, 6]. A way of reducing cost is to genotype individuals from base generations (parents) in HD, and their more numerous descendants at low density (LowD) [6, 7]. Then, using selected SNP from the HD panel, called tagSNP, the non-typed SNP are imputed with high accuracy [7]. Imputing HD genotypes of progeny from LowD genotypes, conditional on grandparental and parental HD genotypes, may result in higher imputation accuracies than those obtained using a reference panel from unrelated individuals [7, 8, 9]. This is because HD genotypes from base generations can be traced within family by means of co-segregation or descendant probabilities [6] while searching for the phase of parental alleles [7].

Most studies on genotype imputation of livestock species have been performed with purebreds [4, 7, 9, 10, 11, 12, 13], and genotype imputation from crossbreds has been largely absent. With regard to agricultural plant species, studies on genotype imputation have used inbred lines [8], recombinant inbred lines (RILs) in Nested Association Mapping (NAM) designs [14, 15], and Multiparent Advanced Generation Inter-Cross studies (MAGIC) [16]. Genotype imputation has also been employed in human studies of genome-linkage analysis for test association of candidate transcriptional regulators with gene expression [17]; and also in a model organism in biomedical research such as the mouse, imputation of genotypes from crosses of inbred lines was used to identify candidates genes for complex disease [18, 19].

Imputing genotypes in humans, plants, livestock, or model organisms, is similar in the sense that a small number of founding individuals can be genotyped at high density, and the bulk of the mapping population can be genotyped at low density using linkage information. In this paper we focus on imputing F2 individuals from a three generation (F0, F1 and F2) population of Duroc × Pietrain crossbred pigs. The F0 and F1 animals were genotyped in HD (60K). The F0 populations used to map QTL in pigs are typically composed of a small number of animals (in our case, 4 males and 15 females) [1, 20, 21, 22]. As it is expected that few recombinations occur in the first generations, these populations have low resolution to map QTL [23]. However, and for the same reason, there is a potential for attaining high accuracy of imputation. The latter effect can be taken to advantage for imputing HD genotypes from inexpensive LowD F2 genotypes, which subsequently allows combining existing data from experimental populations in a meta-analysis for association. There are several reasons for this strategy to be attractive. First, several of these populations have been recently created [21, 22, 24, 25] and DNA from these animals is available. Second, extensive datasets of phenotypes have been recorded for these populations including for traits that are expensive or difficult to measure, such as the content of intramuscular fat and composition of fatty acids [25], age at puberty in gilts [22], and meat tenderness [26]. Finally, these populations are generally developed from breeds that are divergent for some traits of interest such as fat/lean content, meat quality or reproductive efficiency, take for example: Duroc × Pietrain [1, 21], Duroc × Landrace [24], Duroc × Large-White [25], White-Duroc × Erhualian [22], Meishan × Duroc [27], Berkshire × Duroc [20].

Therefore, it follows that imputation of F2 LowD to HD genotypes with high accuracy would be useful and convenient, providing a cost effective strategy as a first step for association analyses or meta-analyses. Different methods have been employed to select tagSNP in LowD panels. Two of them are: 1) imposing restrictions on the minimum value of linkage disequilibrium (LD) or r t 2 between markers [28], 2) selection of tagSNP that are evenly spaced using the physical distance between markers[4, 11, 12]. In addition, commercial chips are also available with medium density segregating SNP selected from several populations, as for example for bovine [29] and pig [10]. A question arises of how many SNP are needed to attain a high accuracy of imputation for a given F2 population. Another question is whether a specific chip has to be custom designed, or whether current commercially available chips can be used. Finally, it is important to determine whether both the F0 and F1 have to be genotyped at HD, or if just genotyping the F0 is adequate to obtain a high accuracy of imputation in the F2.

The goal of this research was to estimate the accuracy of imputation at HD (60K), from LowD F2 genotypes for a Duroc × Pietrain population, using different genotyping schemes. The strategies were evaluated by means of Monte Carlo simulation, conditional on the genotypes from animals in the first two generations (F0 and F1). In doing so, two methods of tagSNP selection were considered and their results were compared to those obtained from a commercial panel chip (9K). In addition to simulations, accuracy of imputation was evaluated using experimental data, taking advantage of a reduced number of F2 animals that were genotyped at HD.


Linkage disequilibrium and selection of tagSNP

Table 1 displays the number of tagSNP selected with different values of LD in an intermediate size chromosome (SSC12), reflected by the measure r t 2. As the value of r t 2 increases, more tagSNP are selected and IA increases. As an example, when r t 2 = 0.2, 79 tagSNP were selected at an average distance of 0.79 Mb and at an accuracy of 0.970. On the other hand for r t 2 = 0.5, 399 tagSNP were selected, positioned at an average distance of 0.16 Mb with IA being equal to 0.982 (Table 1).
Table 1

Accuracy of imputation using tagSNP selected for different values of r t 2 on chromosome 12


Number of tagSNP

Number of SNP genomewide

Average distance between SNP (Mb)

Imputation accuracy ( IA)


























r t 2 = Threshold of LD in statistical selection, Number of tagSNP = Number of SNP selected for a particular r t 2, Number of SNP Genomewide = The equivalent number of SNP that are needed genome-wide to keep the same average inter-marker distance. Average distance between SNP (Mb) = Average distance between tagSNP selected, Imputation accuracy (IA) = Imputation accuracy of non-tagSNP. Results using simulated data.

Evenly spaced SNP

The IA using tagSNP selected using either LD information or evenly spaced SNP were similar. For example, the IA of non-typed SNP on SSC12 were 0.973 and 0.970, respectively, for 80 evenly spaced SNP as compared with 79 tagSNP selected with r t 2 = 0.2 (Figure 1). Results for other densities of tagSNP were similar (Figure 1). Moreover, evenly spaced tagSNP sets of comparable density across chromosomes yielded similar accuracies. Thus, for example an average inter–marker distance of 2.1 Mb, 140 tagSNP on chromosome 1 and 30 tagSNP on chromosome 12 produced IA of 0.969 and 0.968, respectively (Figure 2). In summary, a minimum of 1,200 evenly spaced tagSNP across the genome (average distance = 2.1 Mb) are needed in this F2 population to attain imputation accuracy IA ≥ 0.97 when the F0 and F1 are genotyped with a SNP60 chip.
Figure 1

Accuracy of imputation (IA) using tagSNP selected using LD information or evenly spaced. Imputation accuracy as a function of number of tagSNP selected: evenly spaced tagSNP panels (green dots), tagSNP-LD panels (red squares). Results shown correspond to simulated data from chromosome 12.

Figure 2

Accuracy of imputation (IA) as a function of average tagSNP distance for evenly spaced panels on chromosomes 1 and 12. Imputation accuracy as a function of the average distance between evenly spaced tagSNP in megabases (Mb) for seven SNP panels on simulated data from chromosome 1 (red squares) and chromosome 12 (blue dots).

Imputed genotypes in experimental F2 animals

9K commercial chip

The values of IA were calculated for two scenarios and for each chromosome, using a 9K SNP list that was developed for producing a commercial LowD panel (GeneSeek, Inc., Lincoln, NE, USA; described in Badke et al. [10]).

Imputation accuracies IA were 0.90 and 0.99 when the F1 was genotyped at low or high density, respectively (Figure 3). In the latter case, although the accuracy was high in all chromosomes (0.99), SNP in some regions were imputed with lower accuracy (Figure 4). High IA in the F2 were obtained across all SNP when the F1 was genotyped at HD (Figure 4a,b). However, when the F1 was genotyped at LowD, IA in F2 individuals decreased along the whole chromosome (Figure 4c,d). A logical question to consider is the following: how much accuracy is gained when including pedigree information, when compared with the use of population-wise LD as the unique source of information? To answer this, the imputation was performed again using as reference panel the genotypes of F0 and F1 animals and the F2 at LowD, but without specifying the pedigree of the F2s. In other words, the F2 animals were assumed unrelated and their parents were unknown. For chromosome 1 the results are displayed in Figure 5. Notice that the average IA in the F2 was equal to 0.90. Therefore, the IA was lower than when the information on relationships was used (0.99, Figure 4a,b). This indicates that the inclusion of HD genotypes from related animals and explicitly specifying paternities greatly increases accuracy of imputation.
Figure 3

Accuracy of imputation (IA) for SNP on 60K chip using the 9K panel as tagSNP. Average accuracy of imputation for each chromosome using experimental data: Blue bars correspond to the case of F0 at high density (HD), F1 and F2 at low density (LowD). Red bars correspond to gain in accuracy of imputation when the F1 is genotyped at HD.

Figure 4

Accuracy of imputation (IA) across chromosomes 1 and 12 under two genotyping scenarios. Generation F0 and F1 at high density, F2 at low density: chromosome 1 (a), chromosome 12 (b). Generation F0 at high density, F1 and F2 at low density: chromosome 1 (c), chromosome 12 (d). The blue line displays a local regression fit of the data. All results were obtained using the experimental data.

Figure 5

Accuracy of imputation (IA) per SNP ignoring pedigree relationship on chromosome 1. Imputation accuracy of experimental data as a function of the chromosomal positions of SNP using information on LD only. Generation F0 and F1 genotyped at high density and F2 genotyped at low density with the relationships between F0, F1 and F2 omitted.

The IA from both genotyping scenarios (Figures 3 and 4) reflect an average drop of 0.1 when the F1 is genotyped at LowD. To gain further insight, the simulated haplotypes of two families were used to calculate accuracy of imputation in each scenario. When the F1 is genotyped at LowD, the results showed that the phase error among the SNP that are not tagSNP increased. This loss of accuracy in determining the SNP phase can be traced back to the F0 generation in which the non-tagSNP are also phased with low accuracy. Furthermore, the proportion of SNP with uncertain phase in the F1 genotyped at HD was 4%, and the ensuing accuracy of haplotyping was 0.97. However, when the F1 was genotyped at LowD the proportion of SNP with uncertain phase increased to 30%, and the corresponding accuracy of haplotyping for the non-tagSNP of F1 genotypes dropped to 0.85. In a further analysis with the F1 generation genotyped at HD and used as a reference population (ignoring F0 genotypes), this resulted in 43% of non-tagSNP with uncertain phase in the F1 at HD, and the haplotyping accuracy was even lower (0.78). These results suggest that, in order to have a high accuracy of imputation for non-tagSNP in F2 genotypes, certainty of the phase in the F1 genotypes is required. Such accurately estimated phase is guaranteed when two generations of HD genotypes (F0, F1) are available.

A closer look at Figures 4 and 5 indicates that the position of the SNP had some effect over IA. Therefore, we investigated the relationship between single SNP imputation accuracy and each SNP’s MAF, distance to the nearest tagSNP, and allelic frequency difference between founding breeds.

Minor allele frequency (MAF)

The measure of accuracy based on counting the number of alleles correctly imputed is sensitive to the allelic frequency [8, 12, 30]. In the current study, the square of the correlation (R 2 ) between observed and imputed genotypes was used as a robust measure of accuracy of imputation. It is worth noting that the scale of this measure is somewhat different from the one derived from AI (Table 2).
Table 2

Imputation accuracy of SNP on chromosome 12 measured by IA or by R 2


Genotype design

Accuracy of imputation





R 2













Comparison of two measures of accuracy of imputation calculated using experimental data (tagSNP panel = 30 SNP), under two situations: 1) F0 and F1 at high density (HD), F2 at low density (LowD), 2) F0 at HD, F1 and F2 at LowD.

MAF using the 9K panel in the F2

Figure 6 shows that the MAF of the imputed SNP was not related to R 2 in these data. Notice also that alleles with extreme frequencies (MAF < 0.1) can be imputed with accuracy similar to those SNP at intermediate frequencies (MAF > 0.3).
Figure 6

Imputation accuracy (R 2 ) for SNP on chromosome 12 as a function of the minor allele frequency in the F 0 . Accuracy of imputation of experimental data as a function of minor allele frequency of each SNP (blue dots). Local regression fit (red dots).

Distance to the closest tagSNP

No differences in R 2 were found for the range of distances between non-tagSNP and tagSNP observed (average was equal to 0.936 Mb). Therefore, for an average density between tagSNP of 0.26 Mb, R 2 is similar for a SNP that is in the middle of the interval than for a SNP that is close to the tagSNP (Figure 7). This observation suggests that the density of tagSNP was enough to attain a reasonably equal R 2 for all SNP within the interval.
Figure 7

Imputation accuracy (R 2 ) for SNP on chromosome 12 as a function of the distance to the closest tagSNP. Blue dots are non-tagSNP (Experimental data); distance in base pairs (Log10). Local regression fit is displayed by the red dots.

Effect of the difference in allelic frequencies in the F0

The difference in allelic frequency between founding populations does not seem to affect the R 2 . This means that even SNP that segregate at very different frequencies in founders can be imputed with high accuracy as revealed in Figure 8. Moreover, the apparent drop in R 2 for MAF differences over 0.75 presented in Figure 8 is largely an artefact of very small number of SNP used in the smoothing line fit.
Figure 8

Imputation accuracy (R 2 ) for SNP on chromosome 12 as a function of the difference in allelic frequencies. Accuracy of imputation using experimental data for the difference in allelic frequencies (blue circles) between founding breeds (Pietrain and Duroc). Local regression fit (red dots).


SNP selection methods and accuracy of imputation

A main goal of the present research was to evaluate accuracy of imputation in an F2 cross of pigs (Duroc × Pietrain) using different genotyping scenarios. In a first stage, IA was calculated from simulated F2 data. An ideal situation for linkage based imputation would be to select SNP equally spaced based on genetic distance, as the possibility of recombination between imputed SNP and tagSNP would be minimal. However, this is not possible in the absence of a high resolution linkage map. Consequently, to position the tagSNP we used two proxies: a) physical spacing, and b) LD-based selection. For our simulated population, the two proxies produced the same results, most likely because it was assumed that 1cM = 1 Mb uniform recombination rate. Therefore, in this simulated population, the average distance between tagSNP throughout the genome proved to be a good indicator of accuracy of imputation (IA), as values greater or equal to 0.97 were obtained using average distances among tagSNP that were less than or equal to 2.1 Mb. Next, the selection of tagSNP using the LD method was compared to choosing SNP located at regularly spaced intervals throughout the genome. In the first method, LD was measured by r t 2, the minimum threshold of r2 between any non-tagSNP with at least one tagSNP. It was observed that when r t 2 increased, the number of selected tagSNP and IA also increased. The accuracy was between 0.960 (r t 2 = 0.1) and 0.982 (r t 2 = 0.5), with average distance between tagSNP of 1.86 Mb and 0.16 Mb, respectively. Xu et al. [28] used r t 2 = 0.8 to select a set of tagSNP for genome-wide association analyses in humans. Their use was slightly different from ours in that they were selecting SNP to tag causative variants for genome-wide association using population level LD information only. On the other side, we wanted to use this method to select SNP that were more evenly spaced in terms of genetics distance as done previously with outbred pig populations [10], but this time exploiting within and between family LD. Consequently, low levels of r t 2 were used in the current study as we found that with a threshold of r t 2 ≥ 0.6, many tagSNP were selected with marginal increases of IA. The second method employed to select tagSNP consisted of dividing the chromosome into segments of equal size, and then choosing the SNP that lay closest to the center of the segment. Other studies have used evenly spaced tagSNP by selecting one SNP every given number of markers [12], or by choosing in each segment the SNP with the largest MAF [4],11,12]. The fact that we had available a sizable number of SNP throughout the genome, i.e. 60 K, made it possible to select approximately evenly spaced SNP with a wide range of MAF, as long as those SNP were segregating in the population. The values of IA calculated while using tagSNP chosen at evenly spaced segments were similar to those obtained using the LD method. This similarity of results may be due to an assumption made in the method of SNP selection at evenly spaced intervals, i.e. that the distribution of LD along the genome is almost uniform and there are no large blocks of LD. In the current research, the haplotypes of F1 animals are sampled from two populations: Duroc and Pietrain. The resulting LD was relatively high and uniformly distributed, except for a few blocks with extremely high LD: blocks with at least 7 consecutive SNP with r2 ≥ 0.8. For this reason, evenly spaced tagSNP and tagSNP selected based on the LD method produced similar imputation accuracy at equivalent density. Although we indeed simulated assuming uniform recombination rates, these results seem to agree also with experimental data, where the two methods of selection used here produced virtually the same accuracy in an outbred pig population [10]. Designing custom low density SNP panels for each population of interest would not be cost effective. Consequently, we investigated the imputation accuracy obtained using a commercially available SNP chip with markers selected based on physical position and MAF [10].

Imputation using 9K panel and genotyping scenarios

Data from a 9K chip (average distance between SNP = 0.30 Mb) were used as a LowD panel to impute to a HD 60K panel. Using the experimental data from F2 individuals, different genotyping scenarios were tested. In the first scenario, data consisted of F0 and F1 genotypes at HD and F2 at LowD, and average IA was 0.99. Similarly, Weigel et al. [13] imputed 8K genotypes to 43K using information of the sire, dam, and grandsires (paternal and maternal), and obtained a value of IA > 0.95.

Our second scenario included the F1 genotyped at 9K, between the generations of grandparents and grandoffspring, and it was observed that IA of F2 decreased to 0.9. In our last scenario F0 and F1 were genotyped at HD and F2 at LowD but the relationships between the F2 and the reference panel were ignored, resulting in an average accuracy of imputation of 0.9. Badke et al. [10] used the genotypes of a reference population formed by trios to impute genotypes of an unrelated population, and obtained values of IA of 0.90 and 0.95 using reference groups of 16 and 64 animals, respectively.

Habier et al. [6] indicated that the reasons for the decay in accuracy of imputation are two-fold: 1) the accuracy of haplotyping the tagSNP flanking the non-tagSNP; 2) the accuracy of haplotyping the imputed non-tagSNP, conditional on a correct haplotyping of the tag-SNP. Therefore, the impact of both factors under the first two scenarios and taking into account the relationships between the individuals in the F2 and in the reference population, were evaluated by means of simulated data. Accuracies of haplotyping were calculated as the number of erroneous inference of phase between consecutive heterozygous markers, as in Druet and Georges [31]. In all scenarios, it was observed that the phases of tagSNP were correct, thus the uncertainty was due to the grandparental origin of the non-tagSNP that were flanked by the tagSNP. The next step was to quantify the fraction of non-tagSNP with uncertain phase. When F0 and F1 were genotyped at HD and F2 was genotyped at LowD, the fraction of non-tagSNP with uncertain phase was 4%, whereas this statistic was 30% when the F0 was genotyped at HD, and the F1 and F2 were genotyped at LowD. The corresponding IA were 0.97 and 0.85, respectively. These results suggest that accuracies of imputation in the current study were affected by knowledge of the phase of non-tagSNP. Moreover, when the amount of genotypes from related individuals (i.e., F0 at HD) increases, the accuracy of haplotyping goes up, and the accuracy of imputation also increases. These results apply to genotyping designs with a pedigree with a small number of founder individuals genotyped in HD and a large number of progeny genotyped in LowD. If the phase is known in the founders, it is easy to accurately follow transmission of chromosomal segments to the remainder of the population using linkage information. In practice, however the phase needs to be ascertained using LD information. Such information is very limited in cases such as our F0 because of reduced sample size. In that case, the researchers can follow two paths. First, as presented with large pedigrees, having extra animals from the same founding population(s) can help in using LD to accurately phase those animals. Second, as presented here, two consecutive generations can be genotyped in HD to use the information in grand-parents (F0) to accurately phase the parents (F1) and then use linkage information to impute genotypes within the progeny (F2). For such approaches to work, full pedigree information (three generations) and two generations of HD genotypes are needed. The approach is still cost effective in typical F2 populations [6, 32]. These results are partially reaffirmed in large pedigree based imputation.

MAF effect

The measures of accuracy of imputation that are based only on allelic counts are not useful for comparing SNP having different values of MAF. This is due to the fact that imputation errors are highly sensitive to the value of the allelic frequencies [8, 12, 30]. To overcome this restriction, two alternative measures of accuracy of imputation have been proposed: 1) the correlation between imputed and observed genotypes [8]; and 2) an accuracy of imputation corrected to its expected value [12, 30]. The second method consists of adjusting the calculated accuracy of imputation by the difference between the observed accuracy and an estimate of the expected value under random sampling. There are several possible ways of calculating the accuracy under this method. Regardless of the measure being used to calculate the accuracy, a trend for the accuracy of imputation to drop when MAF < 0.15 has been observed. For example, in maize Hickey et al. [8] observed a decrease in R 2 when MAF < 0.10, and the drop was higher when the masked genotypes were >84% of total SNP. Similarly, Lin et al. [30] used human data with the correction for expected accuracy and observed a marked decrease in accuracy of imputation when MAF < 0.15. Hayes et al. [12] used the same correction as Lin et al. [30] with sheep data and found highly variable accuracies of imputation but tending to decrease whenever MAF < 0.10. The correlation between observed and imputed genotypes (R 2 ) was employed in the current research to evaluate the effect of MAF on imputation accuracy. Our results showed that markers with MAF < 0.10 in the founders were imputed with reasonably good accuracy in the F2 (Figure 6), a result different from those previously discussed. This is not unexpected considering we used both LD and linkage (pedigree information), as sources of information from our crossbred population. Therefore, the allele frequency in the F0 does not matter as long as in that generation the two alleles are segregating. Moreover, whenever the F1 is genotyped at HD, SNP with low MAF can be observed in the F0 and F1. Coupled to the fact that all family relationships are known, this simplifies the imputation of F2 animals.

Possible effects in association

In the current research we compared allelic dosage of observed and imputed genotypes to find accurate genotyping design and imputation methods for LowD genotypes in an F2 population. Zhen et al. [33] reported that the regression of phenotype on allelic dosage was an accurate method to evaluate QTL effects. Moreover, they observed that when accuracies of imputation were high, the power for the association test was high. For example, accuracies of imputation > 0.95 were associated with values of power > 0.85. In the current study, the accuracy of imputation obtained with the 9K panel was R 2 = 0.94, which suggests that the power for an association test is high. Other studies also found that imputation improved the power for association tests. Using data from humans, Hao et al. [34] compared the power for GWAS analysis of four different strategies involving imputation: (1) directly testing for associations using the Illumina 317K SNPs, (2) testing for associations using the entire imputed HapMap SNP set based on the Illumina 317K genotype data; (3) directly testing for associations using the Illumina 650Y SNPs; and (4) testing for associations using the entire imputed HapMap SNP set based on Illumina 650Y genotype data. It was observed that genomic wide imputation (strategies 2 and 4) improved power by 5.5% for the Illumina 317K, or 3.3% for Illumina 650Y, compared to the analyses with assayed SNPs only (strategies 1 and 3, respectively). Similar results were obtained by Anderson et al. [5] for the 300K and 550K platforms.

The cost of genotyping is an important consideration. At present, the cost of commercial HD genotyping (60K) for pigs is more than twice as much as the cost of genotyping with the 9K chip. Assuming a population with a structure similar to the one used here (approximately 20 F0, 56 F1 and 1000 F2), one can genotype 1.9 times more individuals in a scenario with F0 and F1 at HD, and F2 at LowD than in a scenario with F0, F1 and F2 at HD. The imputed genotypes can then be used for association or for meta-analysis studies.


Designing custom SNP panels for each F2 population to be imputed will likely not be cost effective due to the relatively large number of SNP needed to attain reasonable imputation accuracies, and the high development costs for each SNP panel. In particular, for our population we would need a minimum of M = 1,200 markers with average distance of 2.1 Mb to have IA over 0.97 in the F2. On the other hand, using the 9K panel as tagSNP (LowD) resulted in IA of 0.99 when the F0 and F1 were genotyped at HD and the F2 at LowD. The cost of such genotyping scheme would be less than half the cost of using HD genotypes for all individuals. The correlation between observed and imputed genotypes was high (R 2  = 0.94), so that the power for future association studies would be high. Thus, under a genotyping strategy of high accuracy of imputation (i.e., F0 and F1 at HD, F2 at LowD), information on imputed genotypes from more animals that is similar to that from a HD panel can be obtained at a lower cost. These results apply to the imputation of markers in the SNP60 beadchip, in populations where a small number of founders can be genotyped at HD and phase of parents of imputed animals can be derived with certainty. Translation of LD-based results, on the other hand, are constrained to pig populations showing similar levels of LD as in the founding animals [35].



The experimental population was raised at the Michigan State University Swine Teaching and Research Farm, East Lansing, MI [1]. Parents from the initial generation (F0) were four unrelated Duroc boars mated to 15 Pietrain sows by artificial insemination. From all resulting F1 animals, 50 females and 6 males (progeny of 3 F0 sires) were selected as parents for the F2 generation, by avoiding full or half sib matings. A total of 1,259 F2 piglets were born alive from 142 litters out of 11 farrowing groups. Animal protocols were approved by the Michigan State University All University Committee on Animal Use and Care (AUF# 09/03-114-00).

Genotyping and data editing

DNA was isolated from white blood cells using standard procedures as we have previously described for this population [1]. Quantity and quality of DNA samples were determined using a Qubit fluorometer (Invitrogen by Life Technologies, Carlsbad, CA, USA). The number of genotyped animals was N = 411 (4 F0 Duroc boars, 15 F0 Pietrain sows, 6 F1 males, 50 F1 females and 336 F2 pigs). Genotyping was performed at a commercial laboratory (GeneSeek, a Neogen Company, Lincoln, NE, USA) using the Illumina PorcineSNP60 beadchip [36]. Out of M = 62,163 SNP, 6,422 SNP were eliminated as their physical positions were unknown. Mendelian inconsistencies (≤ 0.01%) were taken as missing genotypes, and 12 animals (1 F1 and 11 F2) with more than 10% of SNP missing were not used in any analysis. By similar consideration, 3,038 SNP were removed from the analyses due to presenting more than 10% missing data. Additionally, 10,139 SNP were excluded as their minor allele frequency (MAF) was below 0.01. These editing policies resulted in a data set comprising 399 pigs with 45,003 SNP per animal. This editing procedure followed that of Badke et al. [35] and the program PLINKv1.07 [37] was used. Additionally, starting with genotypes for F0 and F1 animals, genotypes for 932 F2 animals were simulated conditional on the real pedigree using a gene-dropping model. Simulated genotypes were used to assess alternative tagSNP selection procedures while experimental genotypes on a subset of animals (n = 336) were used to assess imputation accuracy using a SNP list for a 9K commercial chip that has recently been publicly released by GeneSeek Inc. (Lincoln, NE, USA; described in Badke et al. [10]).

Genotype simulation

A stochastic simulation was performed to evaluate two different methods of selecting tagSNP for imputation on the accuracy of the resulting F2 genotypes. The genotypes of 932 F2 animals were simulated using gene-dropping[38] theory, by conditioning on a real pedigree and on the haplotypes of the 55 F1 parents (6 males and 49 females) from the real F2 population. The haplotypes were estimated at a high accuracy from the genotypes of the F1 parents and 19 F0 ancestors (4 Duroc boars and 15 Pietrain sows), using the software MERLIN [39]. The number of recombinations in the F1 haplotypes were drawn from a Poisson distribution with mean equal to the length of the given chromosome in Morgans (M) by assuming 1 Mb = 1 cM [40]. The positions of the recombinations were simulated from a uniform distribution using Haldane’s mapping function [41, 42]. As an example, there were 1,405 SNP on chromosome 12 that were spread over 64.2 Mb, and the ensuing average distance between markers was 0.04573 Mb. By assuming a recombination rate of 1 cM per Mb [38], the number of recombinations in chromosome 12 was drawn from a Poisson distribution with parameter equal to 64.2 / 100 = 0.642. The next step was to assign the resulting gametes carrying these recombinations of the F1 genotypes to their F2 progeny.

TagSNP selection using simulated dataset

Two different methods were used for tagSNP selection: 1) The first one consisted of a statistical search built into the software FESTA [43] and used information on LD [44]. In this method, each SNP was either an element of the tagSNPset, or in LD with an existing element in the tagSNPset, at a value equal or larger than a specified threshold (r t 2) [10]. A minimum level of r t 2 based on pair-wise LD of the F1 haplotypes was selected, so that all SNPs above the chosen threshold were selected as tagSNP. 2) The second method consisted of selecting evenly-spaced markers. The chromosome was divided into k segments of equal length, and then the SNP that was closest to the center of the segment was selected. In cases where there were no SNP lying in a segment, no selection was performed resulting in the number of tagSNP≤k in segments of approximately equal length.

Genotype imputation

For simulated data, F2 genotypes of non-typed markers were imputed using the algorithm of Lander and Green [39] that predicts the non-tagSNP by conditioning on the observed markers. For computational reasons the pedigree was analyzed on a per litter basis. Thus, for each F2 litter, a three generation pedigree was built [45] using the four F0 grandparents, the two F1 parents, and up to a maximum of 10 F2 animals. When the litter had more than 10 progeny, a new “family” was formed with the four F0 grandparents, the two F1 parents and the remaining F2 animals. The resulting “families” were analyzed separately and genotypes were imputed with MERLIN [39]. Breaking the pedigree in this way produces some loss of information, but simulation results (data not shown) suggested that the loss was negligible.

For experimental data, F2 genotypes of non-typed markers were imputed using the algorithm built into the software AlphaImpute [4]. The algorithm implemented in AlphaImpute [4] uses information on population-wide and within family LD and it required certain tuning. In particular, we set the core length parameter to 100, 150, 400 and 600 SNP and the tail parameter haplotype to 300, 400, 600 and 800 SNP, respectively. Likewise, genotype error percentage parameter was set to 0%, so as to obtain a high percentage of alleles under the correct phase [46]. The algorithm was run for the entire pedigree as there was no computing restriction in this case.

Calculation of the accuracy of imputation

Irrespective of data generation (simulation or experimental), the accuracy of genotype imputation in F2 individuals for all methods was evaluated using two different statistics. First, the mean of the difference between observed and imputed allelic dosage was calculated [9, 13] as follows:
IA = 1 1 2 N i N j M i g ^ ij g ^ ij Open image in new window
In this expression, N is the total number of animals imputed, M i represents the number of markers with observed genotype in animal i, gij is the observed (experimental or simulated) allelic dosage in animal i and SNP j, and g ^ ij Open image in new window is the corresponding imputed allelic dosage. Allelic dosage was defined as the number of copies of a reference allele that took values 0, 1 and 2 for homozygous reference, heterozygous and homozygous non-reference, respectively. The second expression used to quantify the imputation accuracy was the square of the correlation between observed and imputed genotypes at each allele, or R2 statistics of Huang et al. [47]. Denoting g ^ ¯ Open image in new window, the average value of the imputed genotypes, and with g ¯ Open image in new window the average value of observed genotypes, the R2 statistics were calculated as follows:
R 2 = i = 1 N g ^ ij g ^ ¯ g ij g ¯ i = 1 N g ^ ij g ^ ¯ 2 i = 1 N g ij g ¯ 2 2 Open image in new window

The statistic is interpreted as a squared correlation coefficient.



This project was supported by Agriculture and Food Research Initiative Competitive Grant no. 2010-65205-20342 from the USDA National Institute of Food and Agriculture, and by funding from the National Pork Board Grant no. 11–042. Partial funding was also provided by the US Pig Genome Coordinator. Computer resources were provided by the Michigan State University High Performance Computing Center (HPCC). JLGD and RJCC were funded by UBACyT 20020100100861 from Universidad de Buenos Aires (Argentina). We acknowledge Yvonne M. Badke (Ph.D. student) and Yijian Huang Ph. D. (Postdoctoral researcher) of Michigan State University for help with programing.

Supplementary material

12863_2012_1102_MOESM1_ESM.pdf (43 kb)
Authors’ original file for figure 1
12863_2012_1102_MOESM2_ESM.pdf (38 kb)
Authors’ original file for figure 2
12863_2012_1102_MOESM3_ESM.jpeg (414 kb)
Authors’ original file for figure 3
12863_2012_1102_MOESM4_ESM.pdf (1.1 mb)
Authors’ original file for figure 4
12863_2012_1102_MOESM5_ESM.pdf (330 kb)
Authors’ original file for figure 5
12863_2012_1102_MOESM6_ESM.jpeg (305 kb)
Authors’ original file for figure 6
12863_2012_1102_MOESM7_ESM.pdf (221 kb)
Authors’ original file for figure 7
12863_2012_1102_MOESM8_ESM.pdf (182 kb)
Authors’ original file for figure 8


  1. 1.
    Edwards DB, Ernst CW, Tempelman RJ, Rosa GJM, Raney NE, Hoge MD, Bates RO: Quantitative trait loci mapping in an F2 Duroc x Pietrain resource population: I. Growth traits. J Anim Sci. 2008, 86: 241-253.CrossRefPubMedGoogle Scholar
  2. 2.
    Haley CS, Knott SA, Elsen JM: Mapping quantitative trait loci in crosses between outbred lines using least squares. Genetics. 1994, 136: 1195-1207.PubMedCentralPubMedGoogle Scholar
  3. 3.
    Choi I, Steibel JP, Bates RO, Raney NE, Rumph JM, Ernst CW: Application of alternative models to identify QTL for growth traits in an F2 Duroc x Pietrain pig resource population. BMC Genet. 2010, 11: 97-PubMedCentralCrossRefPubMedGoogle Scholar
  4. 4.
    Hickey JM, Kinghorn BP, Tier B, van der Werf JH, van der Cleveland MA: A phasing and imputation method for pedigreed populations that results in a single-stage genomic evaluation. Genet Sel Evol. 2012, 44: 9-10.1186/1297-9686-44-9.PubMedCentralCrossRefPubMedGoogle Scholar
  5. 5.
    Anderson CA, Pettersson FH, Barrett JC, Zhuang JJ, Ragoussis J, Cardon LR, Morris AP: Evaluating the Effects of Imputation on the Power, Coverage, and Cost Efficiency of Genome-wide SNP Platforms. Am J Hum Genet. 2008, 83: 112-119. 10.1016/j.ajhg.2008.06.008.PubMedCentralCrossRefPubMedGoogle Scholar
  6. 6.
    Habier D, Fernando RL, Dekkers JCM: Genomic selection using low-density marker panels. Genetics. 2009, 182: 343-353. 10.1534/genetics.108.100289.PubMedCentralCrossRefPubMedGoogle Scholar
  7. 7.
    Huang Y, Hickey JM, Cleveland MA, Maltecca C: Assessment of alternative genotyping strategies to maximize imputation accuracy at minimal cost. Genet Sel Evol. 2012, 44: 25-10.1186/1297-9686-44-25.PubMedCentralCrossRefPubMedGoogle Scholar
  8. 8.
    Hickey JM, Crossa J, Babu R, de los Campos G: Factors affecting the accuracy of genotype imputation in populations from several maize breeding programs. Crop Sci. 2012, 52: 654-663. 10.2135/cropsci2011.07.0358.CrossRefGoogle Scholar
  9. 9.
    Zhang Z, Druet T: Marker imputation with low-density marker panels in Dutch Holstein cattle. J Dairy Sci. 2010, 93: 5487-5494. 10.3168/jds.2010-3501.CrossRefPubMedGoogle Scholar
  10. 10.
    Badke YM, Bates RO, Ernst CW, Schwab C, Fix J, Van Tassell CP, Steibel JP: Methods of tagSNP selection and other variables affecting imputation accuracy in swine. BMC Genet. 2013, AcceptedGoogle Scholar
  11. 11.
    Huang Y, Maltecca C, Cassady JP, Alexander LJ, Snelling WM, Macneil MD: Effects of reduced panel, reference origin, and genetic relationship on imputation of genotypes in Hereford cattle. J Anim Sci. 2012, 59301: 1-17.Google Scholar
  12. 12.
    Hayes BJ, Bowman PJ, Daetwyler HD, Kijas JW, van der Werf JHJ: Accuracy of genotype imputation in sheep breeds. Anim Genet. 2011, 43: 72-80.CrossRefPubMedGoogle Scholar
  13. 13.
    Weigel KA, Tassell CPV, O’Connell JR, VanRaden PM, Wiggans GR: Prediction of unobserved single nucleotide polymorphism genotypes of Jersey cattle using reference panels and population-based imputation algorithms. J Dairy Sci. 2010, 93: 2229-2238. 10.3168/jds.2009-2849.CrossRefPubMedGoogle Scholar
  14. 14.
    Tian F, Bradbury PJ, Brown PJ, Hung H, Sun Q, Flint-Garcia S, Rocheford TR, McMullen MD, Holland JB, Buckler ES: Genome-wide association study of leaf architecture in the maize nested association mapping population. Nat Genet. 2011, 43: 159-162. 10.1038/ng.746.CrossRefPubMedGoogle Scholar
  15. 15.
    Poland JA, Bradbury PJ, Buckler ES, Nelson RJ: Genome-wide nested association mapping of quantitative resistance to northern leaf blight in maize. Proc Natl Acad Sci USA. 2011, 108: 6893-6898. 10.1073/pnas.1010894108.PubMedCentralCrossRefPubMedGoogle Scholar
  16. 16.
    Kover PX, Valdar W, Trakalo J, Scarcelli N, Ehrenreich IM, Purugganan MD, Durrant C, Mott R: A multiparent advanced generation inter-cross to fine-map quantitative traits in Arabidopsis thaliana. PLoS Genet. 2009, 5: e1000551-10.1371/journal.pgen.1000551.PubMedCentralCrossRefPubMedGoogle Scholar
  17. 17.
    Burdick JT, Chen W-M, Abecasis GR, Cheung VG: In silico method for inferring genotypes in pedigrees. Nat Genet. 2006, 38: 1002-1004. 10.1038/ng1863.PubMedCentralCrossRefPubMedGoogle Scholar
  18. 18.
    Bouxsein ML, Uchiyama T, Rosen CJ, Shultz KL, Donahue LR, Turner CH, Sen S, Churchill G, Müller R, Beamer WG: Mapping quantitative trait loci for vertebral trabecular bone volume fraction and microarchitecture in mice. Journal of bone and mineral research: the official journal of the American Society for Bone and Mineral Research. 2004, 19: 587-599.CrossRefGoogle Scholar
  19. 19.
    Leduc MS, Hageman RS, Verdugo RA, Tsaih S-W, Walsh K, Churchill GA, Paigen B: Integration of QTL and bioinformatic tools to identify candidate genes for triglycerides in mice. J Lipid Res. 2011, 52: 1672-1682. 10.1194/jlr.M011130.PubMedCentralCrossRefPubMedGoogle Scholar
  20. 20.
    Stearns TM, Beever JE, Southey BR, Ellis M, Mckeith FK, Rodriguez-Zas SL: Evaluation of approaches to detect quantitative trait loci for growth, carcass, and meat analyses The online version of this article, along with updated information and services, is located on the World Wide Web at: Evaluation of approaches to detect. J Anim Sci. 2005, 83: 1481-1493.PubMedGoogle Scholar
  21. 21.
    Liu G, Jennen DGJ, Tholen E, Juengst H, Kleinwächter T, Hölker M, Tesfaye D, Ün G, Schreinemachers H-J, Murani E, Ponsuksili S, Kim J-J, Schellander K, Wimmers K: A genome scan reveals QTL for growth, fatness, leanness and meat quality in a Duroc-Pietrain resource population. Anim Genet. 2007, 38: 241-252. 10.1111/j.1365-2052.2007.01592.x.CrossRefPubMedGoogle Scholar
  22. 22.
    Yang G, Ren J, Li S, Mao H, Guo Y, Zou Z, Ren D, Ma J, Huang L: Genome-wide identification of QTL for age at puberty in gilts using a large intercross F2 population between White Duroc x Erhualian. Genetics. 2008, 40: 529-539.Google Scholar
  23. 23.
    Mackay TFC: The genetic architecture of quantitative traits. Annu Rev Genet. 2001, 35: 303-339. 10.1146/annurev.genet.35.102401.090633.CrossRefPubMedGoogle Scholar
  24. 24.
    Nonneman D, Lindholm-Perry AK, Shackelford SD, King DA, Wheeler TL, Rohrer GA, Bierman CD, Schneider JF, Miller RK, Zerby H, Moeller SJ: Predictive markers in calpastatin for tenderness in commercial pig populations. J Anim Sci. 2011, 89: 2663-2672. 10.2527/jas.2010-3556.CrossRefPubMedGoogle Scholar
  25. 25.
    Sanchez MP, Iannuccelli N, Basso B, Bidanel J-P, Billon Y, Gandemer G, Gilbert H, Larzul C, Legault C, Riquet J, Milan D, Le Roy P: Identification of QTL with effects on intramuscular fat content and fatty acid composition in a Duroc x Large White cross. BMC Genet. 2007, 8: 55-PubMedCentralCrossRefPubMedGoogle Scholar
  26. 26.
    Meyers SN, Rodriguez-Zas SL, Beever JE: Fine-mapping of a QTL influencing pork tenderness on porcine chromosome 2. BMC Genet. 2007, 8: 69-PubMedCentralCrossRefPubMedGoogle Scholar
  27. 27.
    Sato S, Oyamada Y, Atsuji K, Nade T, Sato S-i, Kobayashi E, Mitsuhashi T, Nirasawa A, Komatsuda Y, Saito S, Terai T, Hayashi T, Sugimoto Y: Quantitative trait loci analysis for growth and carcass traits in a Meishan × Duroc F2 resource population. J Anim Sci. 2003, 81: 2938-2949.PubMedGoogle Scholar
  28. 28.
    Xu Z, Kaplan NL, Taylor JA: TAGster: efficient selection of LD tag SNPs in single or multiple populations. Bioinformatics. 2007, 23: 3254-3255. 10.1093/bioinformatics/btm426.PubMedCentralCrossRefPubMedGoogle Scholar
  29. 29.
    Wiggans GR, Cooper TA, Vanraden PM, Olson KM, Tooker ME: Use of the Illumina Bovine3K BeadChip in dairy genomic evaluation. J Dairy Sci. 2012, 95: 1552-1558. 10.3168/jds.2011-4985.CrossRefPubMedGoogle Scholar
  30. 30.
    Lin P, Hartz SM, Zhang Z, Saccone SF, Wang J, Tischfield JA, Edenberg HJ, Kramer JR, Goate A, Bierut LJ, Rice JP: A new statistic to evaluate imputation reliability. PLoS One. 2010, 5: e9697-10.1371/journal.pone.0009697.PubMedCentralCrossRefPubMedGoogle Scholar
  31. 31.
    Druet T, Georges M: A hidden markov model combining linkage and linkage disequilibrium information for haplotype reconstruction and quantitative trait locus fine mapping. Genetics. 2010, 184: 789-798. 10.1534/genetics.109.108431.PubMedCentralCrossRefPubMedGoogle Scholar
  32. 32.
    Druet T, Farnir FP: Modeling of identity-by-descent processes along a chromosome between haplotypes and their genotyped ancestors. Genetics. 2011, 188: 409-419. 10.1534/genetics.111.127720.PubMedCentralCrossRefPubMedGoogle Scholar
  33. 33.
    Zheng J, Li Y, Abecasis GR, Scheet P: A comparison of approaches to account for uncertainty in analysis of imputed genotypes. Genet Epidemiol. 2011, 35: 102-110. 10.1002/gepi.20552.PubMedCentralCrossRefPubMedGoogle Scholar
  34. 34.
    Hao K, Chudin E, McElwee J, Schadt EE: Accuracy of genome-wide imputation of untyped markers and impacts on statistical power for association studies. BMC Genet. 2009, 10: 27-PubMedCentralCrossRefPubMedGoogle Scholar
  35. 35.
    Badke YM, Bates RO, Ernst CW, Schwab C, Steibel JP: Estimation of linkage disequilibrium in four US pig breeds. BMC Genomics. 2012, 13: 24-10.1186/1471-2164-13-24.PubMedCentralCrossRefPubMedGoogle Scholar
  36. 36.
    Ramos AM, Crooijmans RPMA, Affara NA, Amaral AJ, Archibald AL, Beever JE, Bendixen C, Churcher C, Clark R, Dehais P, Hansen MS, Hedegaard J, Hu Z-L, Kerstens HH, Law AS, Megens H-J, Milan D, Nonneman DJ, Rohrer GA, Rothschild MF, Smith TPL, Schnabel RD, Van Tassell CP, Taylor JF, Wiedmann RT, Schook LB, Groenen MAM: Design of a high density SNP genotyping assay in the pig using SNPs identified and characterized by next generation sequencing technology. PLoS One. 2009, 4: e6524-10.1371/journal.pone.0006524.PubMedCentralCrossRefPubMedGoogle Scholar
  37. 37.
    Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, Maller J, Sklar P, de Bakker PIW, Daly MJ, Sham PC: PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet. 2007, 81: 559-575. 10.1086/519795.PubMedCentralCrossRefPubMedGoogle Scholar
  38. 38.
    Ledur MC, Navarro N, Pérez-Enciso M: Large-scale SNP genotyping in crosses between outbred lines: how useful is it?. Heredity. 2010, 105: 173-182. 10.1038/hdy.2009.149.CrossRefPubMedGoogle Scholar
  39. 39.
    Abecasis GR, Cherny SS, Cookson WO, Cardon LR: Merlin–rapid analysis of dense genetic maps using sparse gene flow trees. Nat Genet. 2002, 30: 97-101. 10.1038/ng786.CrossRefPubMedGoogle Scholar
  40. 40.
    Laval G, Excoffier L: SIMCOAL 2.0: a program to simulate genomic diversity over large recombining regions in a subdivided population with a complex history. Bioinformatics. 2004, 20: 2485-2487. 10.1093/bioinformatics/bth264.CrossRefPubMedGoogle Scholar
  41. 41.
    Cheema J, Dicks J: Computational approaches and software tools for genetic linkage map estimation in plants. Brief Bioinform. 2009, 10: 595-608. 10.1093/bib/bbp045.CrossRefPubMedGoogle Scholar
  42. 42.
    Haldane JBS: The combination of linkage values and the calculation of distances between the loci of linked factors. J Genet. 1919, 8: 299-309.CrossRefGoogle Scholar
  43. 43.
    Qin ZS, Gopalakrishnan S, Abecasis GR: An efficient comprehensive search algorithm for tagSNP selection using linkage disequilibrium criteria. Bioinformatics. 2006, 22: 220-225. 10.1093/bioinformatics/bti762.CrossRefPubMedGoogle Scholar
  44. 44.
    Carlson CS, Eberle MA, Rieder MJ, Yi Q, Kruglyak L, Nickerson DA: Selecting a maximally informative set of single-nucleotide polymorphisms for association analyses using linkage disequilibrium. Am J Hum Genet. 2004, 74: 106-120. 10.1086/381000.PubMedCentralCrossRefPubMedGoogle Scholar
  45. 45.
    Haley S, Elsen JM: Mapping quantitative trait loci in crosses between outbred lines using least squares. Genetics. 1994, 136: 1195-1207.PubMedCentralPubMedGoogle Scholar
  46. 46.
    Hickey JM, Kinghorn BP, Tier B, Wilson JF, Dunstan N, van der Werf JHJ: A combined long-range phasing and long haplotype imputation method to impute phase for SNP genotypes. Genet Sel Evol. 2011, 43: 12-10.1186/1297-9686-43-12.PubMedCentralCrossRefPubMedGoogle Scholar
  47. 47.
    Huang L, Li Y, Singleton AB, Hardy JA, Abecasis G, Rosenberg NA, Scheet P: Genotype-imputation accuracy across worldwide human populations. Am J Hum Genet. 2009, 84: 235-250. 10.1016/j.ajhg.2009.01.013.PubMedCentralCrossRefPubMedGoogle Scholar

Copyright information

© Gualdrón Duarte et al.; licensee BioMed Central Ltd. 2013

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 (, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Authors and Affiliations

  • Jose L Gualdrón Duarte
    • 1
    • 2
  • Ronald O Bates
    • 1
  • Catherine W Ernst
    • 1
  • Nancy E Raney
    • 1
  • Rodolfo JC Cantet
    • 2
  • Juan P Steibel
    • 1
    • 3
  1. 1.Department of Animal ScienceMichigan State UniversityEast LansingUSA
  2. 2.Departamento de Producción AnimalFacultad de Agronomía, UBA-CONICETBuenos AiresArgentina
  3. 3.Department of Fisheries and WildlifeMichigan State UniversityEast LansingUSA

Personalised recommendations