(13)C NMR and electron ionization mass spectrometric data-activity relationship model of estrogen receptor binding

Modeling chemical interaction profiles: I. Spectral data-activity relationship and structure-activity relationship models for inhibitors and non-inhibitors of cytochrome P450 CYP3A4 and CYP2D6 isozymes

An interagency collaboration was established to model chemical interactions that may cause adverse health effects when an exposure to a mixture of chemicals occurs. Many of these chemicals—drugs, pesticides, and environmental pollutant—interact at the level of metabolic biotransformations mediated by cytochrome P450 (CYP) enzymes. In the present work, spectral data-activity relationship (SDAR) and structure-activity relationship (SAR) approaches were used to develop machine-learning classifiers of inhibitors and non-inhibitors of the CYP3A4 and CYP2D6 isozymes. The models were built upon 602 reference pharmaceutical compounds whose interactions have been deduced from clinical data, and 100 additional chemicals that were used to evaluate model performance in an external validation (EV) test. SDAR is an innovative modeling approach that relies on discriminant analysis applied to binned nuclear magnetic resonance (NMR) spectral descriptors. In the present work, both 1D ¹³C and 1D ¹⁵N-NMR spectra were used together in a novel implementation of the SDAR technique. It was found that increasing the binning size of 1D ¹³C-NMR and ¹⁵N-NMR spectra caused an increase in the tenfold cross-validation (CV) performance in terms of both the rate of correct classification and sensitivity. The results of SDAR modeling were verified using SAR. For SAR modeling, a decision forest approach involving from 6 to 17 Mold² descriptors in a tree was used. Average rates of correct classification of SDAR and SAR models in a hundred CV tests were 60% and 61% for CYP3A4, and 62% and 70% for CYP2D6, respectively. The rates of correct classification of SDAR and SAR models in the EV test were 73% and 86% for CYP3A4, and 76% and 90% for CYP2D6, respectively. Thus, both SDAR and SAR methods demonstrated a comparable performance in modeling a large set of structurally diverse data. Based on unique NMR structural descriptors, the new SDAR modeling method complements the existing SAR techniques, providing an independent estimator that can increase confidence in a structure-activity assessment. When modeling was applied to hazardous environmental chemicals, it was found that up to 20% of them may be substrates and up to 10% of them may be inhibitors of the CYP3A4 and CYP2D6 isoforms. The developed models provide a rare opportunity for the environmental health branch of the public health service to extrapolate to hazardous chemicals directly from human clinical data. Therefore, the pharmacological and environmental health branches are both expected to benefit from these reported models.

Classification of estrogen receptor-β ligands on the basis of their binding affinities using support vector machine and linear discriminant analysis

Classification models of estrogen receptor-beta ligands were proposed using linear and nonlinear models. The data set was divided into active and inactive classes on the basis of their binding affinities. The two-class problem (active, inactive) was firstly explored by linear classifier approach, linear discriminant analysis (LDA). In order to get a more accurate prediction model, the nonlinear novel machine learning technique, support vectors machine (SVM), was subsequently used to investigate. The heuristic method (HM) was used to pre-select the whole descriptor sets. The model containing eight descriptors founded by SVM, showed better predictive ability than LDA. The accuracy in prediction for the training, test and overall data sets are 92.9%, 85.8% and 91.4% for SVM, 83.1%, 76.1% and 81.9% for LDA, respectively. The results indicate that SVM can be used as a powerful modeling tool for QSAR studies.

Alternative QSAR models for selected estradiol and cytochrome P450 ligands: comparison between classical, spectroscopic, CoMFA and GRID/GOLPE methods

The performance of the spectroscopic EVA (eigenvalue) and EEVA (electronic eigenvalue) methods was tested with data sets applying coumarin 7-hydroxylation inhibitors (28 compounds) for cytochrome P450 mouse CYP2A5 and human CYP2A6 enzymes and 11ss-, 16a-, and 17a-substituted estradiol derivatives (30 compounds) for the lamb uterine estrogen receptor, and compared with the performance of the classical Hansch-type, CoMFA and GRID/GOLPE methods. Besides the internal predictability, the external predictability of the models was tested with several randomized training and test sets to ensure the validity and reliability of the models. Partial least squares (PLS) regression was employed as a general statistical tool with the EVA and EEVA methods. Some supplementary models were also built using only one PLS component with McGowan's volumes (MgVol and MgVol(2)) as additional descriptors and employing multiple linear regression (MLR) as the modelling tool. In general, both the internal and external performance of the EVA model, and more especially the EEVA model, with one PLS component and MgVol parameters was satisfactory, being either as good as or clearly better than that of the Hansch-type, CoMFA and GRID/GOLPE models.

Building an organ-specific carcinogenic database for SAR analyses

FDA reviewers need a means to rapidly predict organ-specific carcinogenicity to aid in evaluating new chemicals submitted for approval. This research addressed the building of a database to use in developing a predictive model for such an application based on structure-activity relationships (SAR). The Internet availability of the Carcinogenic Potency Database (CPDB) provided a solid foundation on which to base such a model. The addition of molecular structures to the CPDB provided the extra ingredient necessary for SAR analyses. However, the CPDB had to be compressed from a multirecord to a single record per chemical database; multiple records representing each gender, species, route of administration, and organ-specific toxicity had to be summarized into a single record for each study. Multiple studies on a single chemical had to be further reduced based on a hierarchical scheme. Structural cleanup involved removal of all chemicals that would impede the accurate generation of SAR type descriptors from commercial software programs; that is, inorganic chemicals, mixtures, and organometallics were removed. Counterions such as Na, K, sulfates, hydrates, and salts were also removed for structural consistency. Structural modification sometimes resulted in duplicate records that also had to be reduced to a single record based on the hierarchical scheme. The modified database containing 999 chemicals was evaluated for liver-specific carcinogenicity using a variety of analysis techniques. These preliminary analyses all yielded approximately the same results with an overall predictability of about 63%, which was comprised of a sensitivity of about 30% and a specificity of about 77%.

Discriminant Function Analyses of Liver-Specific Carcinogens

The ability to predict organ-specific carcinogenicity would aid FDA reviewers in evaluating new chemical applications. A NCTR liver cancer database (NCTRlcdb) containing 999 compounds has been developed with three sets of descriptors. The NCTRlcdb has Cerius2, Molconn-Z, and (13)C NMR descriptors for each compound. Each compound in the database was assigned a liver cancer or a nonliver cancer classification. Compounds within the NCTRlcdb were evaluated for liver-specific carcinogenicity using partial least squares principal component discriminant function (PLS-DF) modeling. PLS-DF models based on estimated a priori classification probabilities of 0.29 for liver cancer and 0.71 for noncancer yielded an overall predictability of 70.6% which was comprised of a liver cancer sensitivity of 18.8% and a noncancer specificity of 90.8%. PLS-DF models based on equal a priori classification probabilities, 0.50 for liver cancer and 0.5 for noncancer, yielded an overall predictability of 61.0% which was comprised of a liver cancer sensitivity of 50.5% and a noncancer specificity of 65.3%.

Spectroscopic QSAR Methods and Self-Organizing Molecular Field Analysis for Relating Molecular Structure and Estrogenic Activity

The performance of three "spectroscopic" quantitative structure-activity relationship (QSAR) methods (eigenvalue (EVA), electronic eigenvalue (EEVA), and comparative spectra analysis (CoSA)) for relating molecular structure and estrogenic activity are critically evaluated. The methods were tested with respect to the relative binding affinities (RBA) of a diverse set of 36 estrogens previously examined in detail by the comparative molecular field analysis method. The CoSA method with (13)C chemical shifts appears to provide a predictive QSAR model for this data set. EEVA (i.e., molecular orbital energy in this context) is a borderline case, whereas the performances of EVA (i.e., vibrational normal mode) and CoSA with (1)H shifts are substandard and only semiquantitative. The CoSA method with (13)C chemical shifts provides an alternative and supplement to conventional 3D QSAR methods for rationalizing and predicting the estrogenic activity of molecules. If CoSA is to be applied to large data sets, however, it is desirable that the chemical shifts are available from common databases or, alternatively, that they can be estimated with sufficient accuracy using fast prediction schemes. Calculations of NMR chemical shifts by quantum mechanical methods, as in this case study, seem to be too time-consuming at this moment, but the situation is changing rapidly. An inherent shortcoming common to all spectroscopic QSAR methods is that they cannot take the chirality of molecules into account, at least as formulated at present. Moreover, the symmetry of molecules may cause additional problems. There are three pairs of enantiomers and nine symmetric (C(2) or C(2)(v)) molecules present in the data set, so that the predictive ability of full 3D QSAR methods is expected to be better than that of spectroscopic methods. This is demonstrated with SOMFA (self-organizing molecular field analysis). In general, the use of external test sets with randomized data is encouraged as a validation tool in QSAR studies.

(13)C NMR quantitative spectrometric data-activity relationship (QSDAR) models of steroids binding the aromatase enzyme

Five quantitative spectroscopic data-activity relationships (QSDAR) models for 50 steroidal inhibitors binding to aromatase enzyme have been developed based on simulated (13)C nuclear magnetic resonance (NMR) data. Three of the models were based on comparative spectral analysis (CoSA), and the two other models were based on comparative structurally assigned spectral analysis (CoSASA). A CoSA QSDAR model based on five principal components had an explained variance (r(2)) of 0.78 and a leave-one-out (LOO) cross-validated variance (q(2)) of 0.71. A CoSASA model that used the assigned (13)C NMR chemical shifts from a steroidal backbone at five selected positions gave an r(2) of 0.75 and a q(2) of 0.66. The (13)C NMR chemical shifts from atoms in the steroid template position 9, 6, 3, and 7 each had correlations greater than 0.6 with the relative binding activity to the aromatase enzyme. All five QSDAR models had explained and cross-validated variances that were better than the explained and cross-validated variances from a five structural parameter quantitative structure-activity relationship (QSAR) model of the same compounds. QSAR modeling suffers from errors introduced by the assumptions and approximations used in partial charges, dielectric constants, and the molecular alignment process of one structural conformation. One postulated reason that the variances of QSDAR models are better than the QSAR models is that (13)C NMR spectral data, based on quantum mechanical principles, are more reflective of binding than the QSAR model's calculated electrostatic potentials and molecular alignment process. The QSDAR models provide a rapid, simple way to model the steroid inhibitor activity in relation to the aromatase enzyme.

Models of polychlorinated dibenzodioxins, dibenzofurans, and biphenyls binding affinity to the aryl hydrocarbon receptor developed using (13)c NMR data

Quantitative spectroscopic data-activity relationship (QSDAR) models for polychlorinated dibenzofurans (PCDFs), dibenzodioxins (PCDDs), and biphenyls (PCBs) binding to the aryl hydrocarbon receptor (AhR) have been developed based on simulated (13)C nuclear magnetic resonance (NMR) data. All the models were based on multiple linear regression of comparative spectral analysis (CoSA) between compounds. A 1.0 ppm resolution CoSA model for 26 PCDF compounds based on chemical shifts in five bins had an explained variance (r(2)) of 0.93 and a leave-one-out (LOO) cross-validated variance (q(2)) of 0.90. A 2.0 ppm resolution CoSA model for 14 PCDD compounds based on chemical shifts in five bins had an r(2) of 0.91 and a q(2) of 0.81. The 1.0 ppm resolution CoSA model for 12 PCB compounds based on chemical shifts in five bins had an r(2) of 0.87 and a q(2) of 0.45. The models with more compounds had a better q(2) because there are more multiple chemical shift populated bins available on which to base the linear regression. A 1.0 ppm resolution CoSA model for all 52 compounds that was based on chemical shifts in 12 bins had an r(2) of 0.85 and q(2) of 0.71. A canonical variance analysis of the 1.0 ppm CoSA model for all 52 compounds when they were separated into 27 strong binding and 25 weak binding compounds was 98% correct. Conventional quantitative structure-activity relationship (QSAR) modeling suffer from errors introduced by the assumptions and approximations involved in calculated electrostatic potentials and the molecular alignment process. QSDAR modeling is not limited by such errors since electrostatic potential calculations and molecular alignment are not done. The QSDAR models provide a rapid, simple and valid way to model the PCDF, PCDD, and PCB binding activity in relation to the aryl hydrocarbon receptor (AhR).

Developing 13C NMR quantitative spectrometric data-activity relationship (QSDAR) models of steroid binding to the corticosteroid binding globulin

We have developed four quantitative spectrometric data-activity relationship (QSDAR) models for 30 steroids binding to corticosteroid binding globulin, based on comparative spectral analysis (CoSA) of simulated 13C nuclear magnetic resonance (NMR) data. A QSDAR model based on 3 spectral bins had an explained variance (r2) of 0.80 and a cross-validated variance (q2) of 0.78. Another QSDAR model using the 3 atoms from the comparative structurally assigned spectral analysis (CoSASA) of simulated 13C NMR on a steroid backbone template gave an explained variance (r2) of 0.80 and a cross-validated variance (q2) of 0.73. Positions 3 and 14 from the steroid backbone template have correlations with the relative binding activity to corticosteroid binding globulin that are greater than 0.52. The explained correlation and cross-validated correlation of these QSDAR models are as good as previously published quantitative structure-activity relationship (QSAR), self-organizing map (SOM) and electrotopological state (E-state) models. One reason that the cross-validated variance of QSDAR models were as good as the other models is that simulated 13C NMR spectral data are more accurate than the errors introduced by the assumptions and approximations used in calculated electrostatic potentials, E-states, HE-states, and the molecular alignment process of QSAR modeling. The QSDAR models developed provide a rapid, simple way to predict the binding activity of a steroid to corticosteroid binding globulin.