- THIS ARTICLE
-
Abstract
- Full Text (PDF)
- Alert me when this article is cited
- Alert me if a correction is posted
- SERVICES
- Similar articles in this journal
- Similar articles in PubMed
- Alert me to new issues of the journal
- Download to citation manager
- Reprints & Permissions
- CITING ARTICLES
- Citing Articles via HighWire
- Citing Articles via Google Scholar
- GOOGLE SCHOLAR
- Articles by Wang, J.
- Search for Related Content
- PUBMED
- PubMed Citation
- Articles by Wang, J.
An Estimator for Pairwise Relatedness Using Molecular Markers
Jinliang Wangaa Institute of Zoology, Zoological Society of London, London NW1 4RY, United Kingdom
Corresponding author: Jinliang Wang, Regent's Park, London NW1 4RY, United Kingdom., jinliang.wang{at}ioz.ac.uk (E-mail)
Communicating editor: A. H. D. BROWN
 | ABSTRACT |
|---|
 TOP ABSTRACT THE NEW ESTIMATOR OF...COMPARISON AMONG ESTIMATORSDISCUSSIONLITERATURE CITED |
|---|
I propose a new estimator for jointly estimating two-gene and four-gene coefficients of relatedness between individuals from an outbreeding population with data on codominant genetic markers and compare it, by Monte Carlo simulations, to previous ones in precision and accuracy for different distributions of population allele frequencies, numbers of alleles per locus, actual relationships, sample sizes, and proportions of relatives included in samples. In contrast to several previous estimators, the new estimator is well behaved and applies to any number of alleles per locus and any allele frequency distribution. The estimates for two- and four-gene coefficients of relatedness from the new estimator are unbiased irrespective of the sample size and have sampling variances decreasing consistently with an increasing number of alleles per locus to the minimum asymptotic values determined by the variation in identity-by-descent among loci per se, regardless of the actual relationship. The new estimator is also robust for small sample sizes and for unknown relatives being included in samples for estimating allele frequencies. Compared to previous estimators, the new one is generally advantageous, especially for highly polymorphic loci and/or small sample sizes.
GENETIC relationships among individuals within and between populations play a pivotal role in quantitative genetics, conservation genetics, and many areas of evolution and ecology (
Marker-based methods for inferring genetic relationships among individuals can be categorized into two groups. The first group uses a likelihood approach to determine the likelihood of a pair of individuals falling into a particular type of relationship (such as full-sibs, parent-offspring, etc.) given the marker data (e.g.,
Relatedness estimators are especially useful for populations with complex (unknown) pedigrees or with little prior information on population structure. Several moment estimators for two-gene and four-gene relatedness have been proposed (e.g.,
| THE NEW ESTIMATOR OF TWO-GENE AND FOUR-GENE COEFFICIENTS OF RELATEDNESS |
|---|
| TOPABSTRACTTHE NEW ESTIMATOR OF...COMPARISON AMONG ESTIMATORSDISCUSSIONLITERATURE CITED |
|---|
The coefficients of relatedness:
Following previous studies, I consider an outbred diploid population throughout. A pair of individuals (say, x and y) in such a population can be genetically correlated in two ways: a single gene at a locus in x is identical by descent with one in y, or both genes in x are identical by descent with those in y. If the probabilities of occurrences of the first and second events are denoted as 
and 
, respectively, the relatedness, r, between individuals x and y can be expressed as
 |
(1) |
(, , and r are 1, 0, and 0.5, respectively, for parents and offspring; 0.5, 0.25, and 0.5, respectively, for full-sibs; and 0.25, 0, and 0.125, respectively, for half-sibs.
Although , , and r can be estimated for any pair of individuals using genotypic data, I concentrate on the estimation of and r only because these two are used to partition the genetic variance of a quantitative trait (
2G) into the additive (2A) and dominance (2D) components, 
( have a much higher sampling variance than those of and r ( cannot be estimated with reasonable confidence.
The new estimator:
To derive the new estimator, I first determine the joint probabilities of two genotypes given the parameters and (to be estimated) and the allele frequencies of the population (assumed to be known for this section). For a marker locus with n codominant alleles indexed by i, j, k, l = 1, 2, ... , n, there would be n(n + 1)/2 possible genotypes and n(n + 1)[n(n + 1) + 2]/8 possible combinations of genotypes for a pair of individuals. The probabilities of these combinations of genotypes can be derived in terms of , , and allele frequencies, as are listed in Table 1 where the population frequency of allele Ai is denoted as pi. According to their similarity index S, defined as the arithmetic average fraction of genes at a locus in a reference individual (here either x or y) for which there is another gene in the other individual (either y or x) that is identical in state (
. Summing up the probabilities of all the genotype combinations within a category gives the probability of occurrence of that category. The probability of occurrence of category 1, for example, is
 |
(2) |
where 
in which the sum of powers of gene frequency 
for m = 2, 3, and 4. The probabilities for categories 2–4 are
 |
(3) |
where 
and 
;
 |
(4) |
where 
and 
; and
 |
(5) |
respectively. Obviously, we have P1 + P2 + P3 + P4 
1, because a pair of genotypes must fall into one of the four categories. The observed values (denoted as 
i) of Pi (i = 1, 2, 3, or 4) for a pair of genotypes constitute the data; one of the four values is one and the rest are zero. For example, if we observe AiAi-AiAi or AiAj-AiAj, then 
and 
.
|
j
For biallelic loci (n = 2), category 3 is not possible, and only three equations remain to be solved for parameters and . Solving any two of the three dependent equations yields
 |
(6) |
Inserting (6) and (7) into (1) yields
 |
(8) |
Equation 7 for is essentially the same as that of and r are, however, different from those of and as well.
With a multiallelic locus, there are more independent equations than the parameters to be estimated. A weighted least-squares analysis could be used to obtain the estimates of and (and thus r) with the minimum sampling variance. Unfortunately, however, the optimum weights must be obtained from the variance-covariance matrix, which depends on the unknown parameters and . To obtain approximate weights, I follow 
in the absence of prior information on the relationship. With this assumption I can obtain the weighted least-squares estimators of and from (2–5) using an equation-solving software,
 |
(9) |
 |
(10) |
where
 |
(11) |
and b–g are defined in (2–4). Inserting (9) and (10) into (1) yields the estimate for r.
When allele frequencies for the population are unknown but are estimated from a sample of N genes taken at random from the population, the sampling error should be taken into account. Denoting the sample allele frequency for allele Ai as 
i, the observed sums of powers of allele frequencies are 
for m = 2, 3, and 4. Obviously, âm is a biased estimate of am for the population. The expected sums of powers of allele frequencies (denoted as 
m for m = 2, 3, and 4) can be obtained using the exact formulas for the first four moments of allele frequency in
 |
(12) |
 |
(13) |
 |
(14) |
As is shown below, essentially unbiased estimates of relatedness can be obtained with the new estimator using (9–14), irrespective of the sample size N.
In practice, relatedness is estimated using several loci. Although relatedness estimates from unlinked loci in linkage equilibrium are independent, they could be dramatically different in sampling variance and ideally should not be simply averaged to give the overall estimate. The optimum weights among loci are difficult to obtain for any moment estimator, however, because they are functions of the parameters ( and ) being estimated. As a result, the difference in the amount of information among loci was either ignored by using equal weights (e.g.,
The average similarity value for unrelated individuals is 
(
, where U is the sum of 1/ul over loci. The averages of i(i = 1, 2, and 3), am (m = 2, 3, and 4), and a22 weighted by w(l) over loci are used in the new estimator. If allele frequencies are estimated from samples, m and 22 are calculated using (12–14) for each locus and then weighted over loci for use in the estimation.
| COMPARISON AMONG ESTIMATORS |
|---|
| TOPABSTRACTTHE NEW ESTIMATOR OF...COMPARISON AMONG ESTIMATORSDISCUSSIONLITERATURE CITED |
|---|
Several moment estimators for pairwise relatedness have been proposed, which are summarized and compared by
Based on the similarity index, or the sharing of genes between individuals,
 |
(15) |
where Sxy is the similarity index between individuals x and y (as defined in Table 1), and is the average similarity among unrelated individuals due to chance.
QUELLER and GOODNIGHT's (1989) estimator (denoted as QG) for the average relatedness between groups of individuals can also be made to estimate the pairwise r, as shown by
 |
(16) |
where pi is the frequency of allele i. Similarly, 
yx using individual y as the reference can be obtained from (16), replacing Sab by Scd and pa and pb by pc and pd, respectively. The arithmetic average of the reciprocal estimates yx and xy yields the estimate of r between individuals x and y.
For estimating r and simultaneously, . The estimator is not shown here because it is quite complex algebraically and involves the solution for weights of two sets of n(n + 5)/2 linear equations for a single locus with n alleles.
On the basis of probability of a genotype conditional on the reference genotype, allele frequencies pi, and parameters r and ,
 |
(17) |
where indicator variables Sij (i, j = a, b, c, or d) are defined in (16). For a number of L loci, the composite estimates are obtained by weighting single-locus estimates across loci,
 |
(19) |
 |
(20) |
where the weights for locus l,
 |
(21) |
 |
(22) |
are derived by assuming that x and y are unrelated (,x are the sums of weights over the L loci. Using individual y as the reference, the estimates yx and 
yx can be obtained similarly. The arithmetic average of the reciprocal estimates yx and xy (xy and yx) yields the estimate of r () between individuals x and y.
Compared to the other estimators, an attractive property of the new estimator (denoted as W) and the QG and LL estimators is that they do not give estimates of r > 1. Like the other estimators, however, they do yield negative estimates because of sampling error. The new estimator does not generate estimates of > 1 except for biallelic loci (in which case the maximum estimate for is 2.125), in contrast to the LR and R estimators.
In the following, I compare the precision and accuracy of these estimators for different actual relationships and allele frequency distributions. First I assume that the allele frequencies of the population are known, and then I assume that allele frequencies are estimated from samples taken binomially from the population without and with relatives-clusters included in them.
Allele frequencies known:
When the allele frequencies are assumed to be known without error, all the pairwise relatedness estimators are essentially unbiased, regardless of the allele frequencies, number of loci, and the actual relationship (
For all of the estimators except for the LR estimator, the average single-locus sampling variance, calculated as the variance of multilocus estimates divided by the number of loci, is independent of the number of loci. Given the allele frequencies and the actual relationship, I can obtain the probability of each pair of genotypes at a single locus and thus the exact sampling variance of relatedness estimates from each estimator.
For the LR estimator, however, the average single-locus sampling variance changes with the number of loci used in the estimation, because the weights for different loci depend in part on the reference genotypes at the corresponding loci (
A peculiar property of the LR estimator is that the single-locus sampling variances of both r and for related individuals may increase with an increasing number of loci. Extensive simulations show that the phenomenon happens more often with a closer actual relationship and a larger number of alleles per locus (data not shown). Allele frequency distributions also affect the direction (increasing or decreasing) of the changes in single-locus variance with an increasing number of loci, as can be seen from Fig 1 in are 0.27, 0.36, 0.42, 0.49, 0.55, and 0.62, respectively. The total sampling variance always decreases but at a decelerating rate with an increasing number of loci. For nonrelatives, the single-locus sampling variance always decreases to an asymptotic value with an increasing number of loci. The odd behavior of the LR estimator is perhaps due to the weights among loci, which are obtained by assuming r = 0 and = 0 (
|
A bizarre property for QG, R, and LR estimators is that, for some allele frequencies, they are undefined because the denominator becomes zero. For biallelic loci, the QG estimator is undefined when the reference individual is a heterozygote. When the allele frequencies are exactly equal for biallelic loci and the reference genotype is a heterozygote, the LR estimator yields undefined estimates. Some equations in the R estimator are invalid when the allele frequency is either 0.5 or 0.25. Allele frequencies near these values cause erratically very large estimates from the three estimators because the denominator is close to zero. This is paradoxical because an allele frequency of 0.5 gives the maximum heterozygosity (for the allele in question) and thus should yield the best estimation. In the following, I avoid biallelic loci in the QG estimator and ignore those component estimates in the R and LR estimators whenever they "blow up."
With known allele frequencies in proportions 1, 2, 3, ... , n (triangular distribution), the five estimators are compared in precision for different numbers of alleles per locus (Fig 1). For the estimation of r, estimators W, LL, and QG give essentially the same results for nonrelatives (almost indistinguishable in the graph) and very similar results for relatives when the number of alleles is large. If the number of alleles is small (n < 5), however, the QG estimator yields higher sampling variances. For nonrelatives, the LR and R estimators have basically the same precision, which is slightly higher than that of the other three estimators across the range of the number of alleles. For relatives, however, the LR and R estimators result in much larger sampling variances than the other estimators when n is large. The sampling variance of r for either full-sibs or parent-offspring nearly levels off for the LR estimator and is actually increasing for the R estimator, when n increases above the value of 12.
For the estimation of , the W estimator gives slightly worse estimates for nonrelatives, slightly better estimates for parent-offspring, and much better estimates for full-sibs where the expected value of is greater than zero. With full-sibs, the sampling variance for the R estimator begins to increase with n when it is >6.
As is intuitively obvious, an increase in the number of alleles per locus should reduce the sampling variance of relatedness estimates, because alleles identical in state are more reliable as indicators of identity-by-descent. The phenomenon that the sampling variance of relatedness (r or ) for relatives from the R and LR estimators does not decrease consistently with an increasing number of alleles is perhaps caused by the assumptions made in deriving weights in these estimators. Without prior information, it is reasonable to assume r = 0 and = 0 for deriving weights among loci () is greater than zero (Fig 1). The higher the actual relatedness is, the more loss in precision due to the violation of the assumption in deriving weights for the two estimators. The optimum weights are a function of the relationship being estimated. Although one can refine the weights iteratively from a starting prior relationship (say, r = 0 and = 0) and use the values when an appropriate degree of convergence has been reached, the resultant estimates are generally much worse even if one uses a large number of loci (
Simulations were also run to compare these estimators for other distributions of allele frequencies. With each of the n alleles being equal to 1/n, the same trend was found as shown in Fig 1, but the estimators are more similar in precision and the sampling variances for the LR and R estimators decrease monotonically with an increasing number of alleles for any actual relationship. Perhaps it is more realistic to assume a uniform Dirichlet distribution of allele frequency, because that is the distribution for populations at equilibrium under the joint effects of drift and mutation or migration (
|
Allele frequencies estimated from samples:
In practice, the allele frequencies of the population are generally unknown and are estimated from samples. Because of practical constraints, the sample sizes are usually not large enough for the sampling effects to be ignored. This is especially obvious for highly polymorphic loci, because they generally have more rare alleles whose frequencies are subject to larger (relative) sampling errors than common alleles for a given sample size. The sensitivity of different estimators to sample sizes for estimating allele frequencies and relatedness is, therefore, important in practice and should be investigated.
There are several practical complications when allele frequencies are estimated. First, a potentially great source of bias in estimating relatedness between a pair of individuals comes from the inclusion of these particular individuals in estimating allele frequencies (-method based on Taylor expansion (
The biases and single-locus variances of r and estimates obtained from different estimators by Monte Carlo simulations on 10 loci are compared in Fig 3 for nonrelatives, parent-offspring, and full-sibs. Twenty alleles per locus were assumed to be in a triangular distribution of frequency in the population, and samples of various sizes were used to estimate the allele frequencies. Across the ranges of sample sizes and actual relationships, the W, LL, and QG estimators are essentially unbiased, while the LR and R estimators give upward-biased estimates of both r and . The bias of the R estimator is very high for relatives and small sample sizes. Although in all cases the bias of the R estimator decreases with increasing sample size, the decrease is rapid only when the actual relatedness (r or ) is zero (Fig 3A and Fig B, for ).
|
The sampling variances of relatedness estimates from the W, LL, and QG estimators are nearly constant over sample sizes for any actual relationship. For the LR and R estimators, the sampling variances decrease with increasing sample sizes when the actual relatedness (r or ) is zero. With the actual relatedness greater than zero, the sampling variances of the LR and R estimators show little change or even increase slightly with sample sizes in the range of 20–100 sampled individuals. Much larger sample sizes are required for the R estimator to yield sampling variances close to the asymptotic values obtained assuming known allele frequencies (Fig 1). For full-sibs in Fig 3C as an example, the single-locus sampling variances (biases) of r and estimates from the R estimator are 0.73 and 3.24 (0.02 and 0.02), respectively, when 500 individuals are sampled, much smaller than the corresponding values 0.77 and 5.42 (0.10 and 0.11) when 100 individuals are sampled, but still larger than the corresponding values 0.42 and 1.23 (0 and 0) with known allele frequencies (in Fig 1).
From Fig 3, it is clear that the LR and R estimators are more sensitive to sampling than the other estimators. The R estimator is especially susceptible to the sizes of samples used in estimating allele frequencies, and the main cause is the treatment of alleles found only in a particular pair of individuals whose relatedness is being estimated. Other treatments of such alleles, such as including them in estimating allele frequencies, also result in large bias and sampling variance for the R and LR estimators. It seems to be difficult for the two estimators to overcome the problems brought by rare alleles in the sample.
In Fig 3, loci with 20 alleles in a triangular distribution of frequency were used in the estimation. More alleles per locus or/and more leptokurtic distributions of allele frequency would require larger sample sizes for the R and LR estimators to obtain estimates with negligible bias and variance close to the asymptotic values obtained with known allele frequencies.
In the above, samples are assumed to be taken binomially from the population and to include nonrelatives only (except for the particular pair of individuals under estimation of relatedness). Clusters of family members may be included in the samples in practice, introducing further sampling errors in estimating allele frequencies (and thus relatedness) if the actual relationship structure among sampled individuals is not accounted for. To investigate the robustness of different estimators when samples involving relatives are used, I considered different proportions of full-sibs from a single family in the sample. Fig 4 compares the biases and single-locus sampling variances among estimators when different proportions of full-sibs are included in a sample of 100 individuals for estimating allele frequencies. The actual relationship is parent-offspring, and 10 loci with 20 alleles in a triangular distribution of frequency were used in estimating relatedness. As is clear from Fig 4, the R estimator is again very sensitive to relatives-clustered sampling. Its upward biases and sampling variances for both r and estimates increase rapidly with an increasing proportion of full-sibs being involved in the sample. The other estimators are robust, with biases increasing slightly and variances almost constant with an increasing proportion of full-sibs in the sample. The relatedness estimation for other actual relationships (nonrelatives, full-sibs) using various numbers of alleles per locus and distributions of allele frequencies were also considered, with essentially the same conclusion.
|
| DISCUSSION |
|---|
| TOPABSTRACTTHE NEW ESTIMATOR OF...COMPARISON AMONG ESTIMATORSDISCUSSIONLITERATURE CITED |
|---|
When population allele frequencies are known, all the pairwise relatedness estimators yield unbiased estimates. These estimates derived from single loci are, however, highly variable with values frequently falling outside of the true range of (0, 1). The W, QG, and LL estimators do not return estimates >1, but do give values <0. The other estimators can yield estimates either >1 or <0. Although estimators can be developed to constrain the estimates to the correct range of relatedness (e.g.,
The large sampling variance of pairwise relatedness estimates comes from several possible sources ( are 0 for nonrelatives and parent-offspring, 1/8 and 3/16 for full-sibs, and 1/16 and 0 for half-sibs. These are the minimum single-locus sampling variances of estimates that can be realized by an efficient estimator in the most favorable conditions. Indeed, Fig 1 and Fig 2 show that the single-locus sampling variances for r from estimators W, LL, and QG, and for from estimator W are approaching these minimum values asymptotically with an increasing number of alleles (n). The single-locus sampling variances for the LR and R estimators asymptote to the minimum values only when the actual r or is zero. Otherwise, they increase for the R estimator, or reach gradually asymptotic values much larger than the minimum variances for the LR estimator, with an increasing n. The problem for the R and LR estimators is caused by the inappropriate weights assuming null relationships.
Second, a great source of the high sampling variance stems from the variation in identity-in-state for alleles that are not identical by descent. More informative loci would give a lower probability of similarity between individuals due to chance (identity-in-state) rather than relationship (identity-by-descent). As is clear from Fig 1 and Fig 2, the sampling variance of relatedness for any estimator changes dramatically with the number of alleles per locus and their frequency distributions. However, there is no general and simple quantity, such as heterozygosity or number of alleles, that could characterize the amount of information at a locus in estimating pairwise relatedness. For the R estimator, the single-locus sampling variances of r and estimates are 1/(n - 1) and 2/[n(n - 1)], respectively, for nonrelatives, which are determined by n only and irrespective of allele frequencies or heterozygosity ( are invariably inversely proportional to the number of loci (L), they have a variable relationship with n depending on the actual relatedness being estimated and allele frequencies. For a given allele frequency distribution and actual nonzero relatedness, the sampling variances generally decrease faster with increasing n when n is smaller (Fig 1 and Fig 2). With the total number of alleles Ln fixed, therefore, an increase in n decreases the sampling variances to a greater extent than an increase in L when n is small. While an increase in L decreases both sampling variance components due to variation in identity-by-descent and in identity-in-state, respectively, an increase in n only reduces the variance caused by the variation in identity-in-state. For actual relationships with large variance in identity-by-descent among loci such as full-sibs, it is more desirable to increase L except when n is very small.
Third, an additional source of single-locus sampling variance for the LR estimator is the number of loci used in the estimation. The problems of the LR estimator are that with an increasing number of loci the average single-locus sampling variance may either increase or decrease, depending on allele frequency distribution and actual relationship, and that the variance may not be able to reach an asymptotic value with a realistically large number of loci, especially for highly polymorphic loci. The cause of the bizarre behavior is the weights among loci, which are derived under the assumption of null relationship.
Fourth, there are differences in sampling variances among statistical methods for estimating relatedness (Fig 1 and Fig 2). The difference is especially large when the distribution of allele frequencies is leptokurtic, the number of alleles per locus is large, and the actual relatedness is greater than zero. The larger sampling variances for relatives from the R and LR estimators are caused mainly by the inappropriate weights derived assuming null relatedness. The difficulty with the R and LR estimators is that the optimum weights that result in the maximum precision are a function of the actual relationship being estimated (
70 diallelic loci (p = 0.5) are necessary for the likelihood estimator to have a similar precision to that of moment estimators (
When allele frequencies of the population are unknown and estimated from samples, more problems arise with the moment estimators. These problems are especially evident with highly polymorphic loci and uneven allele frequency distributions because of the involvement of rare alleles. I have shown that the W, LL, and QG estimators give essentially unbiased estimates with sampling variances that are almost constant, irrespective of sample sizes used for estimating allele frequency. The R and LR estimators are, however, more sensitive to sample sizes in both accuracy and precision. The upward bias and sampling variance for both r and from the R estimator are large (Fig 3) and decrease very slowly with increasing sample size when the actual relatedness is greater than zero. With a realistic sample size and highly polymorphic loci, the R and LR estimators could be worse than the other estimators even in the case of nonrelatives (Fig 3A).
Except for the R estimator, all the moment estimators are robust to clusters of close relatives included in samples for estimating allele frequencies (Fig 4). Even with 40% of the individuals in a sample being full-sibs from a single family, the moment estimators (except for the R estimator) give essentially the same estimates (in terms of bias and sampling variance) as those in the ideal case of no relatives in the sample. This is comforting because moment estimators do not allow for estimating allele frequencies and relatedness simultaneously and samples in practice may include clusters of relatives.
The advantage of the new estimator over LR and R estimators increases with increasing degrees of actual relatedness. Throughout I considered the extreme cases of r = 0 (nonrelatives) and r = 0.5 (parent-offspring, full-sibs). Higher relatedness is possible, but should be found rarely in natural outbreeding populations. With very low degrees of actual relatedness, the LR estimator could become the best if the sample size is very large and the loci are not highly polymorphic. In almost all other cases, however, the new estimator is the most efficient for estimating r and jointly. In current practice, highly polymorphic markers such as microsatellites are preferred in genetic studies, which could have many alleles at a locus (53 alleles at a locus was reported in a great tit population; simultaneously.
| ACKNOWLEDGMENTS |
|---|
I thank Andrew Bourke, Rob Hammond, Bill Jordan, Michael Lynch, Kermit Ritland, and three anonymous referees for helpful comments on an earlier version of this article.
Manuscript received June 22, 2001; Accepted for publication December 12, 2001.
| LITERATURE CITED |
|---|
| TOPABSTRACTTHE NEW ESTIMATOR OF...COMPARISON AMONG ESTIMATORSDISCUSSIONLITERATURE CITED |
|---|
CROW, J. F., and M. KIMURA, 1970 An Introduction to Population Genetics Theory. Harper & Row, New York.
FALCONER, D. S., and T. F. C. MACKAY, 1996 Introduction to Quantitative Genetics, Ed. 4. Longman, Harlow, UK.
GOODNIGHT, K. F. and D. C. QUELLER, 1999 Computer software for performing likelihood tests of pedigree relationship using genetic markers. Mol. Ecol. 8:1231-1234[Medline].
LI, C. C., D. E. WEEKS, and A. CHAKRAVARTI, 1993 Similarity of DNA fingerprints due to chance and relatedness. Hum. Hered. 43:45-52[Medline].
LOISELLE, B. A., V. L. SORK, J. NASON, and C. GRAHAM, 1995 Spatial genetic structure of a tropical understory shrub, Psychotria officinalis (Rubiaceae). Am. J. Bot. 82:1420-1425.
LYNCH, M., 1988 Estimation of relatedness by DNA fingerprinting. Mol. Biol. Evol. 5:584-599[Abstract].
LYNCH, M. and K. RITLAND, 1999 Estimation of pairwise relatedness with molecular markers. Genetics 152:1753-1766
LYNCH, M., and J. B. WALSH, 1998 Genetics and Analysis of Quantitative Traits. Sinauer Associates, Sunderland, MA.
MARSHALL, T. C., J. SLATE, L. E. B. KRUUK, and J. M. PEMBERTON, 1998 Statistical confidence for likelihood-based paternity inference in natural populations. Mol. Ecol. 7:639-655[Medline].
MOUSSEAU, T. A., K. RITLAND, and D. D. HEATH, 1998 A novel method for estimating heritability using molecular markers. Heredity 80:218-224.
QUELLER, D. C. and K. F. GOODNIGHT, 1989 Estimating relatedness using molecular markers. Evolution 43:258-275.
RITLAND, K., 1996a A marker-based method for inferences about quantitative inheritance in natural populations. Evolution 50:1062-1073.
RITLAND, K., 1996b Estimators for pairwise relatedness and inbreeding coefficients. Genet. Res. 67:175-186.
RITLAND, K., 2000 Marker-inferred relatedness as a tool for detecting heritability in nature. Mol. Ecol. 9:1195-1204[Medline].
THOMPSON, E. A., 1975 The estimation of pairwise relationships. Ann. Hum. Genet. 39:173-188[Medline].
THOMPSON, E. A., 1976 A restriction on the space of genetic relationships. Ann. Hum. Genet. 40:201-204[Medline].
THOMAS, S. C. and W. G. HILL, 2000 Estimating quantitative genetic parameters using sibships reconstructed from marker data. Genetics 155:1961-1972
VAN DE CASTEELE, T., P. GALBUSERA, and E. MATTHYSEN, 2001 A comparison of microsatellite-based pairwise relatedness estimates. Mol. Ecol. 10:1539-1549[Medline].
WEIR, B. S., 1996 Genetic Data Analysis II. Sinauer Associates, Sunderland, MA.
WRIGHT, S., 1951 The genetical structure of populations. Ann. Eugen. 15:323-354.
This article has been cited by other articles:
 |
|
 |
|
  T. Shikano Estimation of Quantitative Genetic Parameters Using Marker-Inferred Relatedness in Japanese Flounder: A Case Study of Upward Bias J. Hered., March 1, 2008; 99(2): 94 - 104. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. M. Kang, N. A. Zaitlen, C. M. Wade, A. Kirby, D. Heckerman, M. J. Daly, and E. Eskin Efficient Control of Population Structure in Model Organism Association Mapping Genetics, March 1, 2008; 178(3): 1709 - 1723. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. D. Anderson and B. S. Weir A Maximum-Likelihood Method for the Estimation of Pairwise Relatedness in Structured Populations Genetics, May 1, 2007; 176(1): 421 - 440. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Malosetti, C. G. van der Linden, B. Vosman, and F. A. van Eeuwijk A Mixed-Model Approach to Association Mapping Using Pedigree Information With an Illustration of Resistance to Phytophthora infestans in Potato Genetics, February 1, 2007; 175(2): 879 - 889. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. Csillery, T. Johnson, D. Beraldi, T. Clutton-Brock, D. Coltman, B. Hansson, G. Spong, and J. M. Pemberton Performance of Marker-Based Relatedness Estimators in Natural Populations of Outbred Vertebrates Genetics, August 1, 2006; 173(4): 2091 - 2101. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. A. Oliehoek, J. J. Windig, J. A. M. van Arendonk, and P. Bijma Estimating Relatedness Between Individuals in General Populations With a Focus on Their Use in Conservation Programs Genetics, May 1, 2006; 173(1): 483 - 496. [Abstract] [Full Text] [PDF]
|
|
|
|
 |