Plant & Animal Genomes XIII Conference Plant & Animal Genomes XIII ConferenceJanuary 15-19, 2005
Town & Country Convention Center
San Diego, CA
Gregor Bindler1 , Florian Martin1 , Louis Renaud1 , Etienne Kaelin1 , Paolo Donini1 , Ferruccio Gadani1 , Luca Rossi2
A method for discrimination of tobacco types and varieties was developed using microsatellites markers. A set of 200 polymorphic SSRs markers were developed in tobacco and a statistical tool was developed to identify the most robust markers and to select a marker subset that would best discriminate between Burley, Flue-cured and Oriental tobacco types and the 60 selected varieties of a reference panel. The following statistical analyses and steps were performed.
The first step was to eliminate redundancy among the markers. For that purpose the symmetric uncertainty coefficient, U, (link to entropy and Kullback-Leibler information) was used to define a dissimilarity between the markers in the following way: ä(M,M’)=1-U(M,M’). A non-metric multidimensional scaling method enabled us to represent the dissimilarity matrix and to select a subset of markers. For a given marker selection, the DISQUAL approach was applied. In order to make the model robust, a Multiple Correspondence Analysis leading to a lower dimensional representation space was performed, following which a non-parametric discriminant model was built.
To reduce the number of markers in practical applications of type/variety discrimination, an exhaustive approach was adopted. DISQUAL was applied with all the subsets of the pre-selected markers. For type discrimination, several models based on 3 markers lead to a crossvalidation error rate smaller than 5%. For variety discrimination, several models based on 9 markers lead to a crossvalidation error rate smaller than 7.5%. Results are presented to illustrate the use of SSR markers and statistical models for tobacco type and variety discrimination.