A Multiplatform Chemometric Approach to Modeling of Mosquito Repellents

■183

Figure 9.6: The molecular structures of the pairs of compounds classified as the most sim-

ilar according to the results of HCA.

The number of principal components is determined based on Eigenvalues. If Eigen-

value of a PC is greater than 1, it can be considered in further analysis. If there is more

than two PCs that fulfill this rule, the distribution of the compounds can be studied in all

possible combinations of chosen PCs. Practical application of PCA in the analysis of repel-

lence activity or structural properties of the compounds with repellence activity is actually

in their possibility to point out which sub-group of the compounds from the original group

is possibly separated and based on which feature. This can be important for further selec-

tion of target compounds. Compared to simple HCA, the PCA has significant advantage

since it provides the information which variables are actually responsible for the grouping

of compounds in the space of the chosen PCs, while in a simple HCA this cannot be seen.

However, the double dendrograms that in one graph take into account both the clustering

of compounds and clustering of variables can provide this answer.

9.3.5.3

Sum of ranking differences

Sum of ranking differences (SRD) is a non-parametric method used for compari-

son of different samples, techniques or models. This methods is introduced by Héberger

(Héberger, 2010) and it has become much utilized in many QSAR studied (Kovaˇcevi´c et al.,

2018a). Briefly, the SRD analysis is based on calculation of the rank differences between

the rank of the model and the rank of the reference (Kollár-Hunek and Héberger, 2011). It

is crucial to set the “golden standard” correctly in order to have realistic referent ranking.

SRD can be validated by the comparison of ranks by random numbers (CRRN) procedure.

Detailed theoretical background and working principles of SRD method were introduced