Simulating Kernel Lot Sampling: the Effect of Heterogeneity on the Detection of GMO Contaminations

Guidelines defining kernel sampling strategies for quality analyses have been provisionally adopted for the detection of genetically modified (GM) contamination in kernel lots. However, these guidelines are not specific for GM material detection and are not intended for the sampling of non-uniform distributions, a probable situation with respect to the presence of GM material in kernel lots. An analysis of the problem of non-random distribution, through the investigation of the effectiveness of different sampling techniques in producing representative bulk samples, is presented. The analysis is based on a two-step modelling procedure: 1) the kernel lot is created and 2) the lot is sampled to produce a bulk sample. This allows the identification of optimal sampling techniques as a function of specific combinations of population characteristics. For each of 5 levels of GM impurity, varying between 0.1% and 2%, we investigated the effect of 5 levels of stratification (lot size=10 7 kernels). Our results indicate: I) For every GM level, the higher the heterogeneity level, the more unstable the GM estimate becomes; even modest levels of stratification affect the stability of GM estimates. 2) As the number of increment samples increases, the coefficient of variation (CV) of the estimate decreases. Although the pattern of decrease remains similar across stratification levels, the estimated CV changes: with low levels of stratification. 50 samples are enough to obtain estimates with CV<10%. In the case of modest levels of stratification even 100 samples are not sufficient to maintain CV<10%. In case of strongly heterogeneous lots estimates based on 100 units have CVs around 50%. At the same time, the likelihood of false negative results increases significantly.