Genetic isolation among populations can be effectively investigated by multilocus DNA fingerprinting. If populations have diverged, it is expected that the mean proportion of bands shared by individuals from the same population, B-w, exceeds the corresponding mean, B-b, calculated from pairs of individuals from distinct populations. A problem arises in deciding whether any difference between B-w and B-b is statistically significant. In fact, any two band-sharing data (b(ij)), contributing to B-w or B-b, are not independent if they share a common individual (like b(ij) and b(jl)) This prevents a correct application of parametric tests, such as the Student's t-test. Recently, a modification of this test has been proposed that should avoid the independence problem. Using a large number of samples of fingerprints, simulated from an appropriate 'genetic' model, under a wide range of conditions, we compared the performances of the Student's t-test, the modified t-test and five new permutation tests, where individuals, rather than b(ij) values, are permuted. We found that: (i) the Student's t-test can be very permissive, rejecting too often the null hypothesis when true, but is correct or conservative in certain cases; (ii) the modified t-test is extremely conservative when the null hypothesis is true and very inefficient otherwise; (iii) all five permutation tests are strictly correct, provided that individuals are ordered randomly on gels; and (iv) in this case, the permutation tests are equally efficient, and are not inferior to the Student's t-test when the latter is approximately correct and provides a fair benchmark.
DNA fingerprinting data and the analysis of population genetic structure by comparing band-sharing patterns
BERTORELLE, Giorgio;
1999
Abstract
Genetic isolation among populations can be effectively investigated by multilocus DNA fingerprinting. If populations have diverged, it is expected that the mean proportion of bands shared by individuals from the same population, B-w, exceeds the corresponding mean, B-b, calculated from pairs of individuals from distinct populations. A problem arises in deciding whether any difference between B-w and B-b is statistically significant. In fact, any two band-sharing data (b(ij)), contributing to B-w or B-b, are not independent if they share a common individual (like b(ij) and b(jl)) This prevents a correct application of parametric tests, such as the Student's t-test. Recently, a modification of this test has been proposed that should avoid the independence problem. Using a large number of samples of fingerprints, simulated from an appropriate 'genetic' model, under a wide range of conditions, we compared the performances of the Student's t-test, the modified t-test and five new permutation tests, where individuals, rather than b(ij) values, are permuted. We found that: (i) the Student's t-test can be very permissive, rejecting too often the null hypothesis when true, but is correct or conservative in certain cases; (ii) the modified t-test is extremely conservative when the null hypothesis is true and very inefficient otherwise; (iii) all five permutation tests are strictly correct, provided that individuals are ordered randomly on gels; and (iv) in this case, the permutation tests are equally efficient, and are not inferior to the Student's t-test when the latter is approximately correct and provides a fair benchmark.I documenti in SFERA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
