Multivariate classification of accumulation curves. Application in sugarcane selection

RUEDA CALDERÓN, A.; OSTENGO, S.; BRUNO, C.; BALZARINI, M.

Victoria

Conferencia; XXVIIIth International Biometric Conference; 2016

The aim of this work was to cluster sucrose accumulation curves as a function of their functional parameters. Sucrose data used (Pol%cane) of 9 sugarcane genotypes cultivated in multiple environments of Tucuman were recorded fortnightly during 5 harvest months (5 to 49 environments/genotype). Non-linear regression models were adjusted for 175 accumulation curves and four parameters were estimated: α: sucrose content at the onset of harvest, β1: sugar accumulation rate during early harvest (highest accumulation period), γ: time point at which a significant change in accumulation rate occurs, and β2: accumulation rate in late harvest. Curves were classified using UPGMA hierarchical cluster with Euclidean distance and non-hierarchical cluster (K-means). For these clustering methods, the number of clusters was determined using 26 indices (library Nbclust in R) and the majority rule. In addition, the curves were classified with a Self-Organizing Map (SOM), which determined the number of clusters via a screen plot. Validation of internal classification was performed for all algorithms using three indices: Connectivity, Silhouette width and Dunn index (library clValid in R). Using Mixed Linear Models, we estimated the variance components for genotype (G), environment (E) and GxE interaction, with respect to total variability observed in each parameter of the maturity curve, for each group. UPGMA was most sensitive in identifying differences between curves, forming the highest number of clusters; however, they were highly unbalanced due to the presence of atypical curves. Except for the latter curves, internal validation suggested, for all algorithms, the convenience of classifying accumulation curves into two groups: G1 (n=127 curves; α=14,5; β1=0,54; γ=3,5; β2=0,20) and G2 (n=39 curves; α=12,4; β1=1,24; γ=2,12; β2=0,36). In both clusters, the genotypic effect was high for α (45%), the environment effect was the most important for explaining the variability in β1 and γ (>90%) in β2 (77%); environment was the only parameter for which G×E was significant (11%). The results provide useful information for orienting genetic selection for maturity traits in sugarcane based on multi-environment assessment.