PREDICTION OF RATE CONSTANTS FOR METHYL RADICAL ADDITIONS ON
VINYL TYPE ALKENES USING PARTIAL LEAST SQUARES
K. Héberger
Institute of Chemistry, Chemical Research Center,
Hungarian Academy of Sciences
H-1525 Budapest, P.O. Box 17, Hungary
Rate constants are good candidates for prediction, however, it is not
known, which reactant should be chosen for structure description? Methyl
radical was fixed, then, either the alkene or the adduct radical (product)
is suitable for structure description.
The reaction can be given as:
CH3. + CH2=CXY ®
R-CH2-C.XY
{where CH2=CXY is the vinyl type alkene and R-CH2-C.XY
is the adduct radical;
X, Y are substituents: H, Ph, Me, CN, Cl, COOMe, OCOMe, OMe, CHO, etc.}
Structure entry and modeling
Initial 3D modeling of alkenes and adduct radicals was performed using
Hyperchem (Hypercube Inc. Waterloo, ON). 3D structure was refined by AM1
(semiempirical MO program).
Descriptor generation and reduction
Predictor variables (descriptors) were generated using the DRAGON
program package (Milano Chemometrics).
Objective (non-supervised) variable selection: Pair-wise correlations
were investigated and one of any two descriptors with R > 0.99999
were removed.
From among 200 descriptors 189 (case A) and 181 (case B)
remained.
Principal aim is to build a linear PLS model for prediction purposes. Many
descriptors are easy to interpret. We would like to know, which structural
feature is responsible for the magnitude of the rate constants.
A set of 20 rate constants was characterized using 189
descriptors:
constitutional {(23) e.g. number of various atoms and bonds,
Mw, etc.},
topological {(68) connectivity indices, information content,
path counts, Wiener index, etc.},
Weighted Holistic Invariant Molecular (WHIM) descriptors,
empirical descriptors (3).
Conclusions
-
A reliable QSAR model has been developed for estimation of logarithm of
the rate constant for methyl radical additions on vinyl type alkenes.
-
Root mean square error of prediction of 0.2 for logarithm of rate
constant (~ less than a factor of two in rate constants)
can be achieved using linear PLS models.
-
Proper variable selection (using PCM ranking, genetic algorithm, ANN, etc.)
leads to informative variables suitable for MLR.
-
The rate constant of methyl addition basically depends on electronic states
and shapes of alkenes, the number of multiple bonds and the symmetry in
alkenes: