Molecular Structural Characterizations of Chiral Drug and the Studies of Quantitative Structure-Activity/Property Relationships
|Keywords||Vector of molecular electronegativity distance Projective matrix of atomic interaction field Quantitative structure?property/activity relationships Chiral Stereomers(Stereoisomers)|
Molecular structural characterization is an important and indispensable technique in modern drug molecular design and evaluation of pharmacological profiles. At present, molecular field descriptors, such as physicochemical parameters, quantum mechanical descriptor, energetic descriptors and two-dimensional topological descriptors, can not be used directly to distinguish stereomers (stereoisomers). Stereomers quite often exhibit different chemical and physical properties as well as different biological activities. Computational chemists are now facing the challenge of developing models to predict the beneficial as well as adverse effects of stereomers. In such situations quantitative stereochemical structure-activity relationship modeling (QSSAR) approach is necessary rather than simple quantitative structure-activity/property relationship (QSAR/QSPR).Based on the interactions between different atomic types, molecular electronegative distance for chirality (Vmedc), a novel vector of molecular electronegative distance (Vmed) has been defined and generalized in order to further codify chemical structural information for stereomers. Some QSARs/QSPRs for 3 chiral molecular datasets codified by Vmedc have been modeled to testify the capability distinguishing the stereomers. In drug design, candidate compounds are often designed on the same pharmacophore or fragment structure. Enlightened by this idea, based on the reference plane constructed by those atoms in pharmacophore or fragment structure, the atom classification scheme and the vector of atomic interaction, a novel method of molecular structural characterization, the projective matrix of atomic Interaction field (Pmaif) was defined. Through investigating the resolutions of Pmaif for the structures of various organic molecules as well as drugs, specially for molecules with stereo-structural characterization, and creating QSAR/QSPR, Pmaif was proved to be a type of simple and rapid structural descriptors which could deal with the molecular structures of a number of compounds and had comparative qualities with the three?dimensional descriptors with good resolution for the structures and high relationship ability. Pmaif approach provides a powerful alternative QSAR technique for organic compounds.The main contents and some conclusions are as follows:(1). Two descriptors were proposed to characterize the stereo-structural characterization of organic compounds. A new method for robust variables with significance assessment, best variable subset algorithm (BVS), was proposed according to the assessment of the statistical significance in the cross validation. The method of regression or classification combined with BVS was used to evaluat resolution abilities of two descriptors for various molecular structures. Employing regression or classification technique of multiple linear analysis (MLR) and support vector machine (SVM), some approaches about the selection of variables, the test of statistic significance, the evaluation of quality for a model in quantitative vectorial descriptor?property including physical, chemical, or biological property relationships were proposed. Using partical least squares (PLS) regression technique, some approaches about the selection of optimal principal components, the analysis of relation between the loading vector and original descriptos of Vmedc and Pmaif, the construct of the customary regression model, and the evaluation of quality in the development of quantitative structure?property or activity relationships were developed. The ananlysis of correlation for the principal components of Pmaif with physical, chemical, or biological properties were used to extract the optimum principal components (PCs).(2). Pmaif based on various atomic types, atomic attributes and the projective district size gives the relationship between the projective pattern and different chemical and physical properties as well as different biological activities. One key to implement the Pmaif algorithm with VC++ is to search ASP values of atoms whose values are correlative with substructural characterization. The back-tracking algorithm was used to search the substructure bonded with any one atom. Pmaif can be automatically calculated according to the three dimensional structural characterization of organic compound using the analysis system of molecular matix (ASMM).(3). 31 steroid benchmarks were used to evaluate the ability of the codification for the molecular structures when Pmaif was used. The atom classification schemes and different numbers of projective districts (NPD) were optimized, where the used atom classification schems are HVmed4, HVmedSP, HVmed13 and HEState, respectively and the numbers of NPD were 2, 4, 8 and 16, respectively. When the optimal atom classification scheme and NPD for Pmaif with three types of weak interaction were HVmedSP and 16, respectively, the square correlation coefficients of callback(cross-validation) in the training set (R2/Rcv2) and the mean square error in test set (RMSTs) for the QSAR model using PLS algorithm were R2=0.998(Rcv2=0.877) and RMSTs=0.873, respectively. The strong influencing factors was the steric interaction (R2=0.980, Rcv2=0.837 and RMSTs=0.873), then the electrostatic interaction (R2=0.806, Rcv2=0.763 and 0.903), and the hydrophobic interaction (R2=0.913, Rcv2=0.789 and RMSTs=2.354) was the weakest. Good results obtained here are compared to those obtained results using other chiral descriptors, the method of Pmaif can well present the molecular structural characterization and is a powerful alternative approach to codify organic compound.(4). The excellent results were achieved for the classification analysis when the Vmedc and Pmaif were used to codify 32 ACE inhibitors with 5 chiral carbons, where the topological structures of these compounds were the same. When the descriptors of Vmedc were used, the optimal model was the model using best variable subset SVM classification algorithm (BVSSVMc). In the optimal model, Accuracy in training set (ACCTs) was 95.7%, Accuracy for cross-validation in training set (ACCcv) was 95.7% and Accuracy in test set (ACCTs) was 100%. When the descriptors of Pmaif were used, the optimal atom classification scheme was HVmedSP, the optimal NPD was 8, and the accuracy rates in training cross validating and testing were 100% using the forward stepwise linear discrimination model with four principal compounds (PC1, PC6, PC8 and PC14) selected. These results showed that the descriptors of Vmedc and Pmaif can be well present the molecular stereo-structural characterization and are powerful alternative approachs to codify the organic compounds with stereo-structural characterization.(5). The regression analyses were used to further test the regression performance for descriptors of both Vmedc and Pmaif to express two molecular datasets with the stereo-strucutural characterization, 7 pairs of chiral HPP derivatives and 18 chiral acids. When the descriptos of Vmedc were used, the R2(Rcv2) for 7 pairs of chiral HPP derivatives in the optimal QSAR model using best variable subset mutillinear regression algorithm (BVSMLR) was 0.928(0.861), and the R2(Rcv2) for 18 chiral acids in the optimal QSAR model using BVSSVM regression algorithm was 0.756(0.601). When the descriptors of Pmaif were used, the R2(Rcv2) for 7 pairs of chiral HPP derivatives in the optimal QSAR model using BVSSVM regression algorithm was 0.992(0.968), where the optimal atom classification scheme and NPD were HVmed13 and 2, respectively, and The R2(Rcv2) for 18 chiral acids in the optimal QSAR model using BVSSVM regression algorithm was 0.708(0.610), where the optimal atom classification scheme and NPD were HEState and 2, respectively.(6). The descriptors of Pmaif were further used to codify two chiral molecular data sets, nicotinic and muscarinic acetylcholine receptor agonists and antagonists (3,8-diazabicyclo[4.2.0]octane derivatives and 1,3-oxathiolanes), in QSAR using PLS algorithm. The R2(Rcv2) for four types of the affinities of 14 pairs of chiral 1,3-oxathiolane derivatives in the optimal QSAR model were 0.915(0.765)、0.946(0.930)、0.953(0.855) and 0.946(0.831), respectively, and RMSTs were 0.717, 0.941, 0.621 and 0.721, respectively, where the optimal atom classification scheme was HVmedSP, the optimal NPD was 2 and PC was 15 related to Pmaif descriptors. These values of affinities were strong correlation with the steric and electrostatic interaction. The R2(Rcv2) for the observed and predicted values of pEC50 for 3,8-diazabicyclo[4.2.0]octane derivatives withα4β2 andα3β4 subtypes in the optimal QSAR model were 0.743(0.586) and 0.756(0.568), RMSTs were 0.512 and 0.600, where the optimal atom classification scheme was EState, the optimal NPD was 4, and PC was 5 related to Pmaif descriptors. The values of PEC50 were strong correlation with the steric interaction.