Premium
Predicting toxic equivalence factors from 13 C nuclear magnetic resonance spectra for dioxins, furans, and polychlorinated biphenyls using linear and nonlinear pattern recognition methods
Author(s) -
Buzatu Dan A.,
Beger Richard D.,
Wilkes Jon G.,
Lay Jackson O.
Publication year - 2004
Publication title -
environmental toxicology and chemistry
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.1
H-Index - 171
eISSN - 1552-8618
pISSN - 0730-7268
DOI - 10.1897/02-516
Subject(s) - loo , chemistry , quantitative structure–activity relationship , linear regression , biological system , equivalence (formal languages) , molecular descriptor , artificial neural network , molecule , nonlinear system , linear model , pattern recognition (psychology) , computational chemistry , artificial intelligence , mathematics , stereochemistry , statistics , computer science , organic chemistry , physics , pure mathematics , quantum mechanics , biology
Two quantitative spectrometric data‐activity relationships (QSDAR) models have been developed relating 29 dioxin or dioxin‐like molecules to their toxic equivalence factors (TEFs). These models were based on patterns in simulated 13 C nuclear magnetic resonance (NMR) data with the patterns defined by comparative spectral analysis (CoSA). Two versions of CoSA multiple linear regression (MLR) models using 7 or 10 spectral bins had, respectively, explained variances ( r 2 ) of 0.88 and 0.95, and leave‐one‐out (LOO) cross‐validated variances ( q 2 ) of 0.78 and 0.88. A third, artificial neural network model—using a feed forward, back propagating, three‐layer neural network—produced an r 2 of 0.99, a LOO q 2 of 0.82, and a leave‐three‐out q 2 of 0.81. A postulated reason that the results of these QSDAR models are better than traditional quantitative structure‐activity relationship (QSAR) models is based on the difference in descriptors rather than on any differences in pattern recognition approach. Results suggest that the 13 C NMR spectral data contain molecular quantum mechanical information more reflective of each molecule's biochemical properties than do the calculated electrostatic potentials and molecular alignment assumptions used in developing QSAR models. The QSDAR models provide a rapid, simple way to model the toxicity of dioxin and dioxin‐like compounds.