CORE Organic II Authentic Food

This is a copy from what we have delivered to the Final report for the CORE Organic II funded Project: During the first part of the project the main task of WP7 have been meeting attendance, discussion on data sampling, production of guidelines and templates for data submission and production of templates for registration of data. During the last part of the project, WP7 have received data from WP3, WP4, WP5 and WP6 for statistiscal analysis to find methods for classification of samples. These data have been analysed and the main results from these statistical analysis were presented at a project meeting in Italy 25 ? 26th November 2014. Based on the variables measured in these data it seems to be possible to construct classification rules for classifying new samples into one of the two groups Organic and Conventional without having a large number of misclassified samples. Of course, these promising results of the classification rules constructed are valid only for the populations of samples where our training samples come from. As an example we have constructed a simple classification rule based on measured values on 12 multielement variables in 36 OrgTrace wheat samples from Denmark (three regions and two years). Using evaluation by leave-one-out crossvalidation this rule classified all 36 samples correctly (24 samples from Organic and 12 samples from Conventional). The analytical methods used, multielements, different metabolomics methods, and isotopes, seems to produce promising results considering the practical use of their variables to construct classification rules that can be used to classify new samples into Conventional or Organic. The properties of the classification rules constructed are valid for the populations of samples where our known training samples come from. For other populations we may probably construct new classification rules based on training samples from those populations. The ideal situation would be to define a population, for example the population consisting of all possible wheat samples from Denmark. Then randomly select some samples from this population as our training samples and some of them as our test samples. Construct classification rules based on the training samples and evaluate them on the test samples from the same population. The results assume that by analysis of the selected variables we are able to classify unknown samples as Organic or Conventional by means of a classification rule made by analysis of known samples from the same population. WP7 are still waiting for data from other WP's. When WP7 have received them the data will be analysed statistically and the results will be utilized for scientific publications.

WP7 (den norske delen av prosjektet) skal gi statistisk konsultasjon og råd til alle WP-ene omkring hvordan data skal organiseres før de brukes i multivariable statistiske analyser. Deltagernes data vil bli sjekket for formelle forhold, korrigert og struk turert, satt sammen og klargjort i en datafil (database) før de sendes til WP7 for multivariabel statistisk analyse. Prosedyren for å organisere dataene baseres på prinsipper som hovedsaklig er utviklet og anvendt i det foregående Core Organic QACCP prosj ektet (www.coreorganic.org/research) - prosjekt nummer 1855. Alle variablene på et og samme objekt kan være svært korrelerte, som det tas hensyn til ved å bruke multivariable statistiske metoder. Når dataene er strukturert og akkumulert i en passende for m, skal multivariable statistiske analyser anvendes for å studere hensiktene, målsettingene og hypotesene i prosjektet. For å studere korrelasjonene eller avhengighetene mellom variablene skal vi bruke relevante, gode og moderne metoder for å analysere ko rrelasjonsstrukturer. Flere aktuelle metoder vil bli brukt, herunder for eksempel principal component analysis (PCA), soft independent modeling of class analogy (SIMCA), partial least squares discriminant analysis (PLS-DA) og canonical discriminant analys is (CDA). Det kan også være interessant å forsøke å finne spesifikke grupper eller strukturer i dataene. Til det vil vi bruke aktuelle klustrings- og ordinasjonsmetoder.