Assessment of power and accuracy of methods for detection and frequency-estimation of null alleles

Null alleles represent a common artefact of microsatellite-based analyses. Rapid methods for their detection and frequency estimation have been proposed to replace the existing time-consuming laboratory methods. The objective of this paper is to assess the power and accuracy of these statistical tools using both simulated and real datasets. Our results revealed that none of the tests developed to detect null alleles are perfect. However, combining tests allows the detection of null alleles with high confidence. Comparison of the estimators of null allele frequency indicated that those that account for unamplified individuals, such as the Brookfield2 estimator, are more accurate than those that do not. Altogether, the use of statistical tools appeared more appropriate than testing with alternative primers as null alleles often remain undetected following this laborious work. Based on these results, we propose recommendations to detect and correct datasets with null alleles.