model construction. The models are then validated prior to application. If they fail
validation, new descriptors need to be collected and/or new forms of models need to
be generated. Only models that pass external validation can be used to predict or
have any physical meaning attached to them. Details for each step are discussed
below.
Another type of QSSR that has a distinct methodology in generating the descriptors is the molecular interaction field (MIF)-based 3D-QSSR pioneered by
Lipkowitz [58] and Kozlowski [59]. MIF describes the whole 3D structure of the
molecules and therefore is termed a global descriptor. It consists of interaction
energies between a selected chemical probe placed on each point on a threedimensional grid over the 3D structures of the molecules, creating a large amount
of data at a practical computational cost. These data can be strongly correlated so
PLS is a preferred method to generate a 3D-QSSR model. This technique circumvents one of the difficult questions in QSSR model generation about which local
descriptors to collect, since the whole of the information on the 3D structures are
recorded within the MIF. However, the conformations and the alignments of the
molecular structures will strongly affect the values of the MIF so they need to be
handled with careful considerations. A commonly known type of MIF-based QSSR
is the comparative molecular field analysis method or CoMFA. MIF-based QSSRs
have been adopted by many in both academic groups [60–62] and industry [63, 64]
5.2 Collecting Molecular Descriptors
A crucial part of QSSR model building is collecting or generating the
physiochemical descriptors (independent parameters, x i , which describe the chiral
ligands/catalyst/substrates/additive structures). These parameters can take the form
of empirical (i.e. macroscopic) parameters from experiment, such as Hammett
constants, or computed (i.e. atomistic) parameters. Data curation is required for
treatment of empirical data in similar manner to the dependent parameters (Y i ).
Numerous sources of physiochemical descriptors are available in the literature such
as empirical and computed steric parameters: A parameters [65], B parameters [66],
Charton parameters [67], Taft parameters [68], Verloop’s Sterimol parameters [69],
their variants for substituted cyclopentadienyl rings [70] and conformationally
flexible chains [71], Bo’s three-dimensional descriptors [72], electronic parameters,
Hammett parameters, charges, orbital energies, and other parameters such as spectroscopic parameters (IR frequencies, NMR shielding tensors, etc.). There are also
descriptors developed especially for certain types of compounds, e.g. for P-donor
ligands such as Tolman angles [73–75] and continuous chirality descriptors developed by Denmark for chiral compounds [76]. Further treatment of these data by
standardization is sometimes practised (Eq. 3). This transforms the raw data to
dimensionless data with mean ¼ 0 and standard deviation (SD) ¼ 1.
Ligand Design for Asymmetric Catalysis: Combining Mechanistic and. . .
173
validation, new descriptors need to be collected and/or new forms of models need to
be generated. Only models that pass external validation can be used to predict or
have any physical meaning attached to them. Details for each step are discussed
below.
Another type of QSSR that has a distinct methodology in generating the descriptors is the molecular interaction field (MIF)-based 3D-QSSR pioneered by
Lipkowitz [58] and Kozlowski [59]. MIF describes the whole 3D structure of the
molecules and therefore is termed a global descriptor. It consists of interaction
energies between a selected chemical probe placed on each point on a threedimensional grid over the 3D structures of the molecules, creating a large amount
of data at a practical computational cost. These data can be strongly correlated so
PLS is a preferred method to generate a 3D-QSSR model. This technique circumvents one of the difficult questions in QSSR model generation about which local
descriptors to collect, since the whole of the information on the 3D structures are
recorded within the MIF. However, the conformations and the alignments of the
molecular structures will strongly affect the values of the MIF so they need to be
handled with careful considerations. A commonly known type of MIF-based QSSR
is the comparative molecular field analysis method or CoMFA. MIF-based QSSRs
have been adopted by many in both academic groups [60–62] and industry [63, 64]
5.2 Collecting Molecular Descriptors
A crucial part of QSSR model building is collecting or generating the
physiochemical descriptors (independent parameters, x i , which describe the chiral
ligands/catalyst/substrates/additive structures). These parameters can take the form
of empirical (i.e. macroscopic) parameters from experiment, such as Hammett
constants, or computed (i.e. atomistic) parameters. Data curation is required for
treatment of empirical data in similar manner to the dependent parameters (Y i ).
Numerous sources of physiochemical descriptors are available in the literature such
as empirical and computed steric parameters: A parameters [65], B parameters [66],
Charton parameters [67], Taft parameters [68], Verloop’s Sterimol parameters [69],
their variants for substituted cyclopentadienyl rings [70] and conformationally
flexible chains [71], Bo’s three-dimensional descriptors [72], electronic parameters,
Hammett parameters, charges, orbital energies, and other parameters such as spectroscopic parameters (IR frequencies, NMR shielding tensors, etc.). There are also
descriptors developed especially for certain types of compounds, e.g. for P-donor
ligands such as Tolman angles [73–75] and continuous chirality descriptors developed by Denmark for chiral compounds [76]. Further treatment of these data by
standardization is sometimes practised (Eq. 3). This transforms the raw data to
dimensionless data with mean ¼ 0 and standard deviation (SD) ¼ 1.
Ligand Design for Asymmetric Catalysis: Combining Mechanistic and. . .
173