Page 144 - Chiral Separation Techniques

4.7 Decision Tree using Application of Machine Learning 121

As the entire tree is complex and cannot be clearly displayed in one screen, we
report in Fig. 4-17 an expanded (zoomed) fraction of the “nonaromatic” population
set of the tree.

Fig. 4-17. Zoomed picture of the decision tree in the “nonaromatic” region. (0) indicates “no”; (1)
indicates “yes”. (A) for any atom except hydrogen. ($) indicates that the bond is part of a ring, and (!)
bond is part of a chain. CC(C)(C)A for tBu.

Since this current study is restricted to the best enantioseparations (α > 1.8), it is
quite clear that the tree does not accurately reflect the full information contained in
CHIRBASE.
However, it has provided some interesting results. At the top of the tree, the
molecule population is first divided according to the presence or absence of the
attribute “NH2” (primary amine). If the answer is “yes”, the developed branches (on
the right of the tree) mostly leads to the Crownpak CSP. The next attribute is “Aro-
matic”. If the answer is “no”, here the predominant CSP is Chiralpak AD. Aromatic
compounds form the largest part of the tree and as expected the dominant CSP is
Chiralcel OD which is disseminated in almost every region of the tree.