Improved Representation of Biological Information by Using Correlation as Distance Function for Heatmap Cluster Analysis

Abstract

Heatmap cluster figures are often used to represent data sets in the omic sciences. The default option of the frequently used R heatmap function is to cluster data according to Euclidean distance, which groups data mainly to their numerical value and not to its relative behaviour. The disadvantage of using the default clustering dendrograms of R is demonstrated. Instead, a script is provided that uses correlation as distance function, which better reveals biologically meaningful information. This optimized script was used to detect heterotic groups in Vitamaize hybrids (purple maize with high nutraceutical value). A field trial with different genetic combinations was performed through an agricultural phenomics approach (holistic evaluation of the phenotype). The grain yield data and other phenotypic variables were represented through heatmap figures. In the data set of Mexican tropical maize germplasm, at least three heterotic groups were detected, in contrast to only two heterotic groups reported earlier in temperate yellow maize from USA and Europe. This optimized script for heatmap correlation bicluster can also be used to better represent metabolomic fingerprints and transcriptomic data sets.