Fig. 5.
Simulation of different inter-laboratory variabilities and normalization techniques. We simulate the simple case of a single plant analyzed by the 9 laboratories associated with Obépine. Panel (a) shows the results if the WWI normalization formula is applied with a common to all laboratories. Results show a clear disparity between laboratories and a strong attenuation towards laboratories with lower quantification results than laboratory 6. Panel (b) illustrates the correction brought by using a specific to each laboratory. Results are significantly improved for laboratories 4 to 8. The difference is not significant for the remaining 3 laboratories which all have a scaling factor close to 1 and a good inter-samples replicability. Panel (c) shows the correction brought by using ILA results and estimating a scaling factor between each laboratory and Lab 1. As shown in (d), CMILA still is the overall best normalization technique. CM, LSM and CMILA stands for a common maximum, a laboratory-specific maximum and a common maximum after scaling following ILA, respectively. Root mean square errors (RMSE) are calculated using the Lab 1 as reference.