|
Internet Electronic Journal of Molecular Design - IEJMD, ISSN 1538-6414, CODEN IEJMAT
| ABSTRACT - Internet Electron. J. Mol. Des. July 2003, Volume 2, Number 7, 435-453 |
Using Simulated 2D 13C NMR Nearest Neighbor Connectivity Spectral Data Patterns to
Model a Diverse Set of Estrogens
Richard D. Beger, Kathleen J. Holm, Dan A. Buzatu, and Jon G. Wilkes
Internet Electron. J. Mol. Des. 2003, 2, 435-453
|
Abstract:
The scope of this investigation is to develop a rapid, objective modeling method that will
accurately predict the estrogen receptor binding affinities for a diverse set of compounds.
We have used simulated 2D 13C-13C COSY NMR spectral data to develop a model for
130 diverse organic compounds whose relative binding affinities (RBA) to the estrogen
receptor are known. The simulated 2D 13C-13C COSY NMR spectra were generated by
using the NMR spectral assignments for predicted carbon chemical shifts to identify
nearest neighboring carbon atoms and establish carbon-to-carbon through-bond
connectivity spectral patterns of each compound. We call the use of such patterns for
model building comparative structural connectivity spectra analysis (CoSCoSA). For the
large number of estrogens, a CoSCoSA multiple linear regression (MLR) model using 16
bins selected from the 13C-13C COSY
spectral data had an r2 of 0.827, a leave-one-out
cross-validation q12 of 0.78, and a leave-13-out
cross-validation average q132 of 0.78. A
second CoSCoSA model using 15 bins plus one additional distance-related 3D constraint
had an r2 of 0.833, a q12 of 0.79 and
an average q132 of 0.78. The predictions for 27
external compounds had qpred2 of 0.53 for one
CoSCoSA model. The addition of more
through-space distance related 3D information, which presently awaits software
development to make pattern definition for each compound practical, should improve the
predictive accuracy of the CoSCoSA models.
|