Comparison of similarity coefficients used for cluster analysis based on RAPD markers in wild olives

Abstract

ABSTRACT. Five different similarity coefficients (Jaccard, Sorensen-Dice, simple matching, Rogers and Tanimoto, and Russel and Rao) were evaluated and 10 wild olives analyzed with RAPD markers. The influence of the similarity coefficients on wild olives clustering was investigated. Forty-five primers were used on samples from 10 wild olives (Wild 1 and 2 obtained from Mugla province; Wild 3, 4, 5, 6, 7, and 8 from Manisa province and Wild 9 and 10 from Izmir province of Turkey). The similarity matrices obtained from RAPD markers were compared by the Mantel test. Cluster analysis was made with UPGMA dendrograms, and the consensus fork indexes between all pairs of dendrograms were calculated. The Jaccard and Sorensen-Dice coefficients gave the same results, due to the fact that both exclude negative co-occurrences. The dendrograms using the simple matching and Rogers and Tanimoto coefficients were similar; Wild 4 (Akhisar, Manisa) and Wild 9 (Bornova, Izmir) olives had the closest genetic similarities. This occurred because these coefficients include negative co-occurrences. The Russel and Rao coefficients produced different results, because they include negative co-occurrences in the denominator. We concluded that the coefficients that Similarity coefficients for cluster analysis in wild olives do not include negative co-occurrences are more efficient for studies of wild olives clustering based on RAPD markers.

Short Communication Comparison of methods for estimates of molecular genetic diversity in genus Croton: influence of coefficients, clustering strategies and data projection

“…In contrast, the Ward method showed the most unsatisfactory results (0.27 < r c < 0.62; -20379 < D < 32370; 1333 < S < 1703). Similar results were found by Gonçalves et al (2008) in studies with tomato accessions and by Sesli and Yegenoglu (2010) in studies with wild olives. In general, our results support the findings reported by Mohammadi and Prasanna (2003) in that the UPGMA is among agglomerative hierarchical methods most commonly used.…”

“…Among the statistical methods used for diversity studies are i) similarity (and dissimilarity) coefficients, ii) hierarchical grouping methods and grouping methods by optimization, and iii) data projection in two-dimensional space (Duarte et al, 1999;Mohammadi and Prasanna, 2003). In view of the large number of statistical methods available in the literature and the fact that existing comparative studies reveal that the choice of statistical methods can affect the results (Meyer et al, 2004;Gonçalves et al, 2008;Cerqueira-Silva et al, 2009;Sesli and Yegenoglu, 2010;Alves et al, 2012), it is important to establish criteria for choosing statistical methods that are appropriate for the reality of each study to be conducted. M.M.…”

Comparison of statistical methods for genetic similarity evaluation of maize inbred lines

“…Their research was focused predominantly on dominant marker data (RAPD, AFLP) and on coefficients suitable for dichotomic variables (Duarte et. al, 1999;Meyer et al, 2004;Dalirsefat et al, 2009;Sesli et al, 2010). Reif et al (2005) gave detailed analysis of ten various coefficients of similarity/distance applied for germplasm surveys with molecular markers for different purposes in maize breeding pointing out several factors that should be considered prior to decision which one will be used: properties of markers used, genealogy of germplasm, objects under study (lines, populations, etc.…”

Similarity to a single set

“…Jaccard [Jac01] was developed for botany, Ochiai [Och57] was developed for marine zoology (it is also known as Cosine in other areas) and both have been used in many other domains. As well as botany and software engineering, Jaccard has been used in disciplines such as ecology [CCCS05], chemistry [FVB02], genetics [SY10], paleontology [PH99], [LHN06] and psychology [Tve77], to give just a few examples.…”