Bayesian statistical approaches to compositional analyses of transgenic crops 2. Application and validation of...

Bayesian approaches to evaluation of crop composition data allow simpler interpretations than traditional statistical significance tests. An important advantage of Bayesian approaches is that they allow formal incorporation of previously generated data through prior distributions in the analysis steps. This manuscript describes key steps to ensure meaningful and transparent selection and application of informative prior distributions. These include (i) review of previous data in the scientific literature to form the prior distributions, (ii) proper statistical model specification and documentation, (iii) graphical analyses to evaluate the fit of the statistical model to new study data, and (iv) sensitivity analyses to evaluate the robustness of results to the choice of prior distribution. The validity of the prior distribution for any crop component is critical to acceptance of Bayesian approaches to compositional analyses and would be essential for studies conducted in a regulatory setting. Selection and validation of prior distributions for three soybean isoflavones (daidzein, genistein, and glycitein) and two oligosaccharides (raffinose and stachyose) are illustrated in a comparative assessment of data obtained on GM and non-GM soybean seed harvested from replicated field sites at multiple locations in the US during the 2009 growing season.