relative stabilities of stereodetermining transition structures. When using computed
molecular descriptors, as in the context of MLR models, conformational effects
should also be accounted for. Steric and electronic parameters may be sensitive to
conformation, and so it is important to ensure that either the global energy minimum
conformer is used, or preferably a Boltzmann average over all low-energy conformers. A danger of using tabulated values is that the conformer used for parameter
generation is not always readily apparent. Steric parameters dependent on substituent width or length are most sensitive to conformation, which led us to develop an
automated approach to obtain weighted Sterimol parameters (Fig. 16) [71]. Flexible
substituents able to adopt several low-energy conformations with different steric
characteristics contribute to model uncertainty, so that selectivity prediction should
be performed more cautiously in these cases.
5.6 Predicting Yield
Cordeiro and Gonzalez-Diaz have produced a QSPR-perturbation theory to model
the yield or enantioselectivity of intramolecular carbolithiations (Fig. 17)
[107]. Their model was expressed in a complex function of multiple variables
such as variation of chiral ligands, organolithiums and additives, solvents, and
temperatures and time. Rather than giving prediction outcome that converts directly
to % yield or % ee, the model calculated perturbations of reaction outcomes from the
referenced state: λ new ¼ λ reference + perturbation functions, where λ i are binary
functions that take value of 1 or À1 according to a predefined cut-off value. For
example, λ gives a value of 1 for a reaction yield greater than 95% and enantiomeric
excess greater than 90%; otherwise, λ takes value of À1. The authors show that the
model works for a test set. While the interpretation of more complex models is
challenging, the ability to perform multiobjective optimization, in this case of yield
and enantioselectivity, is highly desirable.
Fig. 16 Definition of Sterimol parameters (L, B 1 , B 5 ) and the importance of conformational
sampling, leading to the ensemble average, minimum and maximum values that are used in the
wSterimol approach
184
R. Ardkhean et al.
molecular descriptors, as in the context of MLR models, conformational effects
should also be accounted for. Steric and electronic parameters may be sensitive to
conformation, and so it is important to ensure that either the global energy minimum
conformer is used, or preferably a Boltzmann average over all low-energy conformers. A danger of using tabulated values is that the conformer used for parameter
generation is not always readily apparent. Steric parameters dependent on substituent width or length are most sensitive to conformation, which led us to develop an
automated approach to obtain weighted Sterimol parameters (Fig. 16) [71]. Flexible
substituents able to adopt several low-energy conformations with different steric
characteristics contribute to model uncertainty, so that selectivity prediction should
be performed more cautiously in these cases.
5.6 Predicting Yield
Cordeiro and Gonzalez-Diaz have produced a QSPR-perturbation theory to model
the yield or enantioselectivity of intramolecular carbolithiations (Fig. 17)
[107]. Their model was expressed in a complex function of multiple variables
such as variation of chiral ligands, organolithiums and additives, solvents, and
temperatures and time. Rather than giving prediction outcome that converts directly
to % yield or % ee, the model calculated perturbations of reaction outcomes from the
referenced state: λ new ¼ λ reference + perturbation functions, where λ i are binary
functions that take value of 1 or À1 according to a predefined cut-off value. For
example, λ gives a value of 1 for a reaction yield greater than 95% and enantiomeric
excess greater than 90%; otherwise, λ takes value of À1. The authors show that the
model works for a test set. While the interpretation of more complex models is
challenging, the ability to perform multiobjective optimization, in this case of yield
and enantioselectivity, is highly desirable.
Fig. 16 Definition of Sterimol parameters (L, B 1 , B 5 ) and the importance of conformational
sampling, leading to the ensemble average, minimum and maximum values that are used in the
wSterimol approach
184
R. Ardkhean et al.