Table 2.
Comparison of the predictive ability (Qcum2) and goodness of fit (R2Ycum) of PLS models obtained from GPCR sequences using ACC transformed descriptors with maximum lag 80
| A. Models of 11 receptor classes (prostanoid and orphan receptors not included) | ||||||
| A-component models | 18 component models | |||||
| Centering of z-scales | Normalization (p) | A | R2Ycum | Qcum2 | R2Ycum | Qcum2 |
| 1a original | 1 | 17 | 0.912 | 0.833 | 0.918 | 0.844 |
| 2a original | 0.5 | 16 | 0.909 | 0.866 | 0.922 | 0.879 |
| 3a original | 0 | 16 | 0.905 | 0.841 | 0.920 | 0.858 |
| 4a centred | 1 | 12 | 0.890 | 0.820 | 0.936 | 0.877 |
| 5a centred | 0.5 | 12 | 0.894 | 0.860 | 0.939 | 0.905 |
| 6a centred | 0 | 16 | 0.925 | 0.871 | 0.936 | 0.882 |
| B. Models of 12 receptor classes (prostanoid receptors included). p denotes normalization according to Eq. 1 and 2 | ||||||
| A-component models | 18 component models | |||||
| Centering of z-scales | Normalization (p) | A | R2Ycum | Qcum2 | R2Ycum | Qcum2 |
| A denotes the number of significant PLS components according to cross validation. A-component model represents the PLS model with A number of components. 18-component models represent PLS models with 18 PLS components. | ||||||
| 1b original | 1 | 16 | 0.886 | 0.800 | 0.904 | 0.826 |
| 2b original | 0.5 | 18 | 0.910 | 0.864 | 0.910 | 0.864 |
| 3b original | 0 | 17 | 0.897 | 0.836 | 0.906 | 0.847 |
| 4b centred | 1 | 13 | 0.885 | 0.814 | 0.922 | 0.860 |
| 5b centred | 0.5 | 13 | 0.888 | 0.858 | 0.927 | 0.895 |
| 6b centred | 0 | 17 | 0.918 | 0.872 | 0.925 | 0.880 |