Fig. 2

PCA analysis output from PCM. PCA analysis of the binding site amino acid descriptors corresponding to the 11 mammalian cyclooxygenases considered in the second case study (Proteochemometrics). Binding site amino acid descriptors (5 Z-scales) were input to the function PCA. The first two principal components (PCs) explained more than 80% of the variance. This indicates that there are mainly two sources of variability in the data. To generate the plot, we used the function PCAPlot using the default options. Cyclooxygenases cluster into two distant groups, which correspond to the isoenzyme type, i.e. COX-1 and COX-2. Given that small molecules tend to display similar binding profiles within orthologues [43], we hypothesised that merging bioactivity data from paralogues and orthologues will lead to more predictive PCM models [28].