Correlation analyzes of isoprenoid accumulation - a statistical meta-analysis
All correlation analyses were performed with the aid of R version 3.3.0 (R Core Team, 2016, https://www.R-project.org/) using the outliers (Komsta, 2011, R package version 0.14, https://CRAN.R-project.org/package=outliers) and the gplots (Warnes et al., 2016, R package version 3.0.1, https://CRAN.R-project.org/package=gplots). The significance level α of 0.001 was assumed in all statistical tests.
Although for each accession the level of each metabolite was measured in triplicate, the values thus obtained were analyzed separately, as indicated by the number of experimental points in the respective figures (which equals three times the number of accessions). Means were not calculated, and this approach was employed deliberately to avoid the problem of adjusting and weighing mean values and to allow testing for outliers among single replicates instead of among mean values.
The Shapiro-Wilk test (Shapiro and Wilk 1965) was used to assess the agreement of isoprenoid content in the populations with the Gaussian distribution. Since, even after filtering out of extreme values with the Grubbs’ test for outliers (Grubbs 1950), a vast majority of the distributions were found non-Gaussian, further analyses were performed using non-parametric methods. Consequently, a correlation matrix for the seven investigated isoprenoids was calculated accordingly to the Spearman’s rank correlation coefficients (Spearman 1904).
A hierarchical cluster analysis of the correlation matrix was performed according to the Ward criterion (Ward 1963).