Phenotypic characterization and multivariate analyses

Abstract

The mass utilization of personal computers in agricultural studies in recent times has provided the application of multivariate analysis methods that facilitated recognition of simultaneous interdependence among three or more independent variables. If our data base encompasses records on the phenotypic characterization of the breeding material according to principles of the UPOV descriptor, a simple screening of desired traits by the application of some of multivariate analysis methods, a very important information could be gained in just a couple of minutes regardless of monitoring of a few or a complete set of traits. A set of 58 inbreeds, phenotypically described in one year and one replication according to UPOV descriptor, was analyzed. According to the three estimates (the angle between the blade and stem on leaf just above upper ear, leaf attitude of blade and leaf width), the aim of the study was to determine whether the observed lines were actually natural, homogenous groups on the basis of these three properties. The hierarchical cluster analysis, Ward’s method, was applied, and credibility of results was tested by the discrimination analysis. Based on groups predefined by the cluster analysis, linear combinations of independent variables were formed by the discrimination analysis, hence the error of an incorrect classification was minimal. The first three discrimination functions encompassed 100% of a variance, meaning that the smallest error of the incorrect classification was if the given data set was divided into four groups. The formation of the unique data bases, that become an inexhaustible source of information for experts working on planned and directed breeding, dictates a need to ascertain the efficient way for the utilization of the enormous scope of information wherein the applied methodology in such a work can be of an exceptional importance.

Masovna upotreba PC računara i u agronomskim istraživanjima u novije vreme omogućila je primenu metoda multivarijacione analize koje omogućavaju sagledavanje simultane međizavisnosti između tri ili više nezavisno promenljivih. Ukoliko u svojoj bazi podataka imamo fenotipsku karakterizaciju selekcionog materijala po principima UPOV-og deskriptora, bilo da smo se odlučili za praćenje manjeg broja osobina ili za kompletan set osobina, jednostavnim skriningom željenih osobina uz primenu neke od metoda multivarijacionih analiza možemo za par minuta dobiti važne informacije. Za analizu je uzet set od 58 linija koje su fenotipski opisane u jednoj godini i jednom ponavljanju po principima UPOV-og deskriptora. Posmatrajući tri ocene: ugao stabla i prvog lista iznad klipa, položaj lista i širina lista, cilj istraživanja je bio da se utvrdi da li ispitivane linije prave prirodne, homogene grupe na osnovu ove tri karakteristike. Primenjena je hijerarhijska klaster analiza Ward-sov metod, a verodostojnost rezultata je testirana diskriminacionom analizom. Na osnovu unapred definisanih grupa klaster analizom, diskriminacionom analizom su formirane linearne kombinacije nezavisno promenljivih tako da je greška pogrešne klasifikacije minimalna. Prve tri diskriminacione funkcije obuhvatile su 100% varijanse to znači da je najmanja greška pogrešne klasifikacije onda kada je dati set podataka podeljen u 4 grupe. Formiranje jedinstvenih baza podataka, koje postaju neiscrpan izvor korisnih informacija za stručnjake koji se bave planskim i usmerenim oplemenjivanjem, diktira potrebu iznalaženja efikasnog načina za korišćenje ogromnog obima informacija u čemu korišćena metodologija u datom radu može biti od izuzetnog značaja.