A novel method for improvement of predictability of the CNIBS/R-K equation

A new method for obtaining the model constants of the combined nearly ideal binary solvent/Redlich-Kister (CNIBS/R-K) equation, via least square analysis has been presented. Predictability of CNIBS/R-K in a previous method and the new one of least square analysis has been compared using some experimental solubility data sets. The results have indicated that the new method improved the predictability of the CNIBS/R-K equation about 63%.     

Comparison of various cosolvency models for calculating solute solubility in water-cosolvent mixtures

Previously published cosolvency models are critically evaluated in terms of their ability to mathematically correlate solute solubility in binary solvent mixtures as a function of solvent composition. Computational results show that the accuracy of the models is improved by increasing the number of curve-fit parameters. However, the curve-fit parameters of several models are limited. The combined nearly ideal binary solvent/Redlich-Kister, CNIBS/R-K, was found to be the best solution model in terms of its ability to describe the experimental solubility in mixed solvents. Also resented is an extension of the mixture response surface model. The extension was found to improve the correlational ability of the original model.     

A model to represent solvent effects on the chemical stability of solutes in mixed solvent systems

The applicability of the combined nearly ideal binary solvent/Redlich-Kister (CNIBS/R-K) equation for quantification of solvent effects on the stability of a solute is shown employing the experimental data of three solutes in different aqueous binary solvents. The proposed model provides a simple computational method to correlate/predict the instability rate constant of a drug in mixed solvent systems. The accuracy of the model is compared with that of a model proposed by Connors and co-workers employing various methods including mean percentage deviation (MPD) as comparison criteria. The obtained overall MPD values for the proposed model to correlate and predict the instability rate constants are 2.05 +/- 1.44 and 4.41 +/- 3.21%, respectively, where the corresponding values for Connors' model are 4.34 +/- 3.28 and 10.74 +/- 9.86%. The results suggest that by using only five experimental instability rate constants at different concentrations of the cosolvent in a binary mixture, it is possible to predict unmeasured values falling between data points within an acceptable error range.     

Modeling drug solubility in water-cosolvent mixtures using an artificial neural network

Application of the artificial neural network (ANN) to calculate the solubility of drugs in water-cosolvent mixtures was shown using 35 experimental data sets. The networks employed were feedforward backpropagation errors with one hidden layer. The topology of neural network was optimized and the optimum topology achieved was a 6-5-1 architecture. All data points in each set were used to train the ANN and the solubilities were back-calculated employing the trained networks. The differences between calculated solubilities and experimental values was used as an accuracy criterion and defined as mean percentage deviation (MPD). The overall MPD (OMPD) and its S.D. obtained for 35 data sets was 0.90 +/- 0.65%. To assess the prediction capability of the method, five data points in each set were used as training set and the solubility at other solvent compositions were predicted using trained ANNs whereby the OMPD (+/-S.D.) for this analysis was 9.04 +/- 3.84%. All 496 data points from 35 data sets were used to train a general ANN model, then the solubilities were back-calculated using the trained network and MPD (+/-S.D.) was 24.76 +/- 14.76%. To test the prediction capability of the general ANN model, all data points with odd set numbers from 35 data sets were employed to train the ANN model, the solubility for the even data set numbers were predicted and the OMPD (+/-S.D.) was 55.97 +/- 57.88%. To provide a general ANN model for a given cosolvent, the experimental data points from each binary solvent were used to train ANN and back-calculated solubilities were used to calculate MPD values. The OMPD (+/-S.D.) for five cosolvent systems studied was 2.02 +/- 1.05%. A similar numerical analysis was used to calculate the solubility of structurally related drugs in a given binary solvent and the OMPD (+/-S.D.) was 4.70 +/- 2.02%. ANN model also trained using solubility data from a given drug in different cosolvent mixtures and the OMPD (+/-S.D.) obtained was 3.36 +/- 1.66%. The results for different numerical analyses using ANN were compared with those obtained from the most accurate multiple linear regression model, namely the combined nearly ideal binary solvent/Redlich-Kister equation, and the ANN model showed excellent superiority to the regression model.     

Thermodynamic modeling of activity coefficient and prediction of solubility: Part 1. Predictive models

A new activity coefficient model was developed from excess Gibbs free energy in the form G(ex) = cA(a) x(1)(b)...x(n)(b). The constants of the proposed model were considered to be function of solute and solvent dielectric constants, Hildebrand solubility parameters and specific volumes of solute and solvent molecules. The proposed model obeys the Gibbs-Duhem condition for activity coefficient models. To generalize the model and make it as a purely predictive model without any adjustable parameters, its constants were found using the experimental activity coefficient and physical properties of 20 vapor-liquid systems. The predictive capability of the proposed model was tested by calculating the activity coefficients of 41 binary vapor-liquid equilibrium systems and showed good agreement with the experimental data in comparison with two other predictive models, the UNIFAC and Hildebrand models. The only data used for the prediction of activity coefficients, were dielectric constants, Hildebrand solubility parameters, and specific volumes of the solute and solvent molecules. Furthermore, the proposed model was used to predict the activity coefficient of an organic compound, stearic acid, whose physical properties were available in methanol and 2-butanone. The predicted activity coefficient along with the thermal properties of the stearic acid were used to calculate the solubility of stearic acid in these two solvents and resulted in a better agreement with the experimental data compared to the UNIFAC and Hildebrand predictive models.     

An improved empirical model to calculate solute solubility in supercritical carbon dioxide

To provide more accurate solubility predictions in supercritical carbon dioxide (SC-CO2) using an empirical model employing density as an independent variable, the density of SC-CO2 at different temperatures and pressures has been calculated and compared with experimental densities. The average percentage deviation (APD) has been determined as an accuracy criterion and the obtained APD for the equations studied were between 1.3 (+/-1.4)-11.6 (+/-8.9)%. To show the effects of density values on solubility prediction, the solubility of 18 drug compounds in SC-CO2 has been calculated using an empirical equation with respect to temperature, pressure and density. The APD values for correlative analysis was 8.5 (+/-5.8)% for the most accurate density values calculated by BACK equation of state. A minimum number of experimental data (i.e. 6 points) has been used to train the model then the solubility at other temperatures and pressures has been predicted and the APD value for the most accurate densities obtained was 14.2 (+/-9.4)%. This prediction error could be considered as acceptable when it is compared with RSD values for repeated measurements (approximately 10%) and the proposed predictive method could be employed in industry to calculate the solubility of a drug using a limited number of experimental data.     

Prediction of Apparent Oral Clearance of Small-Molecule Inhibitors in Pediatric Patients

The purpose of this study was to build regression models for the prediction of apparent oral clearance (CL/F) for small-molecule inhibitors in the pediatric population using data obtained from adults. Two approaches were taken; a simple allometric regression model which considers no interdrug or interindividual variability and an allometric regression model with mixed-effects modeling where some variability parameters are included in the model. Average CL/F values were obtained for 15 drugs at various dosages from 31 literatures (a total of 139 data sets) conducted in adults and for 15 drugs from 26 literatures (62 data sets) conducted in children. Data were randomly separated into the “modeling” or “validation” data set, and the 2 allometric regression models were applied to the modeling data set. The predictive ability of the models was examined by comparing the observed and model-predicted CL/F in children using the validation data set. The percentage root mean square error was 17.2% and 26.3% in the simple allometric regression model and the allometric regression model with mixed-effects modeling, respectively. The predictive ability of the 2 models seems acceptable, suggesting that they could be useful for predicting the CL/F of new small-molecule inhibitors and for determining adequate doses in clinical pharmacotherapy for children.     

Prediction of blood-brain distribution: effect of ionization

Two simple multiple linear regression models were proposed to calculate the logarithm of the blood to brain concentration ratio (log BB) of drugs or drug-like compounds. The drugs were classified into two groups according to their ionization state in blood, and the significant parameters were selected using the train sets for each group. For un-ionizable compounds, the logarithm of distribution coefficient in octanol-water in pH 7.4 (log D(7.4)) and molecular weight are the significant parameters, whereas for ionizable compounds, log D(7.4) and number of hydrogen bond acceptor are significant parameters. The developed models were validated and their prediction capabilities checked using an external dataset of 25 compounds. In addition to the acceptable prediction errors, comparison of the external data analysis results with previously proposed models confirmed superior prediction capability of newly developed models.     

Predicting blood-brain barrier penetration of drugs using an artificial neural network

An artificial neural network model is developed to predict the ratios of the steady-state concentrations of drugs in the brain to those in the blood (log BB) from their molecular structural parameters. These molecular structural parameters are the molecular volume (V), the sum of the absolute values of the net atomic charges of oxygen and nitrogen atoms which are hydrogen-bond acceptors (Q(O, N)), and the sum of the net atomic charges of hydrogen atoms attached to oxygen or nitrogen atoms (Q(H)). For a training set of 56 compounds and a test set of 5 compounds, root mean squared errors (RMSE) between experimental log BB values and calculated/predicted log BB values were 0.236 and 0.258, respectively. These molecular structural parameters can be obtained easily from quantum chemical calculations. The model is suitable for the rapid prediction of the blood-brain barrier penetration of drugs.     

人工神经网络预测盐酸帕罗西汀缓释微丸的药物释放

目的利用人工神经网络对盐酸帕罗西汀缓释微丸的释药行为进行预测。方法设计20个处方,其中16个处方作为训练处方,其余4个处方作为测试处方,制备盐酸帕罗西汀膜控释微丸,进行释放度检查。以致孔剂PVPK30的用量、包衣增重作为自变量,考察药物在各个取样点的累积释放量作为输出,建立盐酸帕罗西汀缓释微丸释药行为的人工神经网络预测模型。通过线性回归法、相似因子法、AIC法评价人工神经网络的预测能力。结果通过实测数据和BP神经网络预测结果比较,验证了人工神经网络的预测精度达0.989 9。结论用人工神经网络对盐酸帕罗西汀缓释微丸的释药行为进行预测,拟合度较高,从而为盐酸帕罗西汀缓释微丸的处方优化和释药行为预测提供了可行的依据。     

Prevalidation of a new in vitro reconstituted human cornea model to assess the eye irritating potential of chemicals.

This multicentre study aimed at evaluating the reliability (reproducibility) and relevance (predictivity) of a new commercially available human corneal epithelial (HCE) model (SkinEthic Laboratories, Nice, France) to assess acute ocular irritation. A prevalidation approach (protocol optimisation, transfer and performance) was followed and at each of the four participating laboratories, 20 coded reference chemicals, covering the whole range of irritancy, were tested. The compounds were applied topically to the HCE cultures and the level of cytotoxicity (tissue viability and histological analysis) was determined. Once a standardised protocol was established, a high level of reproducibility between the laboratories was observed. In order to assess the capability of the HCE model to discriminate between irritants (I) and non-irritants (NI), a classification prediction model (PM) was defined based on a viability cut-off value of 60%. The obtained in vitro classifications were compared with different in vivo classifications (e.g. Globally Harmonised System) which were calculated from individual rabbit data described in the ECETOC data bank. Although an overall concordance of 80% was obtained (sensitivity = 100% and specificity = 56%), the predictivity of the HCE model substantially increased when other sources of in vivo and in vitro data were taken into account.