ADME evaluation in drug discovery. 2. Prediction of partition coefficient by atom-additive approach based on atom-weighted solvent accessible surface areas

A novel method for the calculations of 1-octanol/water partition coefficient (log P) of organic molecules has been presented here. The method, SLOGP v1.0, estimates the log P values by summing the contribution of atom-weighted solvent accessible surface areas (SASA) and correction factors. Altogether 100 atom/group types were used to classify atoms with different chemical environments, and two correlation factors were used to consider the intermolecular hydrophobic interactions and intramolecular hydrogen bonds. Coefficient values for 100 atom/group and two correction factors have been derived from a training set of 1850 compounds. The parametrization procedure for different kinds of atoms was performed as follows: first, the atoms in a molecule were defined to different atom/group types based on SMARTS language, and the correction factors were determined by substructure searching; then, SASA for each atom/group type was calculated and added; finally, multivariate linear regression analysis was applied to optimize the hydrophobic parameters for different atom/group types and correction factors in order to reproduce the experimental log P. The correlation based on the training set gives a model with the correlation coefficient (r) of 0.988, the standard deviation (SD) of 0.368 log units, and the absolute unsigned mean error of 0.261. Comparison of various procedures of log P calculations for the external test set of 138 organic compounds demonstrates that our method bears very good accuracy and is comparable or even better than the fragment-based approaches. Moreover, the atom-additive approach based on SASA was compared with the simple atom-additive approach based on the number of atoms. The calculated results show that the atom-additive approach based on SASA gives better predictions than the simple atom-additive one. Due to the connection between the molecular conformation and the molecular surface areas, the atom-additive model based on SASA may be a more universal model for log P estimation especially for large molecules.     

The role of in vitro ADME assays in antimalarial drug discovery and development

The high level of attrition of drugs in clinical development has led pharmaceutical companies to increase the efficiency of their lead identification and development through techniques such as combinatorial chemistry and high-throughput (HTP) screening. Since the major reasons for clinical drug candidate failure other than efficacy are pharmacokinetics and toxicity, attention has been focused on assessing properties such as metabolic stability, drug-drug interactions (DDI), and absorption earlier in the drug discovery process. Animal studies are simply too labor-intensive and expensive to use for evaluating every hit, so it has been necessary to develop and implement higher throughput in vitro ADME screens to manage the large number of compounds of interest. The antimalarial drug development program at the Walter Reed Army Institute of Research, Division of Experimental Therapeutics (WRAIR/ET) has adopted this paradigm in its search for a long-term prophylactic for the prevention of malaria. The overarching goal of this program is to develop new, long half-life, orally bioavailable compounds with potent intrinsic activity against liver- and blood-stage parasites. From the WRAIR HTP antimalarial screen, numerous compounds are regularly identified with potent activity. These hits are now immediately evaluated using a panel of in vitro ADME screens to identify and predict compounds that will meet our specific treatment criteria. In this review, the WRAIR ADME screening program for antimalarial drugs is described as well as how we have implemented it to predict the ADME properties of small molecule for the identification of promising drug candidates.     

An evaluation of thermodynamic models for the prediction of drug and drug-like molecule solubility in organic solvents

Prediction of solubility of active pharmaceutical ingredients (API) in different solvents is one of the main issue for crystallization process design. Experimental determination is not always possible because of the small amount of product available in the early stages of a drug development. Thus, one interesting perspective is the use of thermodynamic models, which are usually employed for predicting the activity coefficients in case of Vapour-Liquid equilibria or Liquid-Liquid equilibria (VLE or LLE). The choice of the best thermodynamic model for Solid-Liquid equilibria (SLE) is not an easy task as most of them are not meant particularly for this. In this paper, several models are tested for the solubility prediction of five drugs or drug-like molecules: Ibuprofen, Acetaminophen, Benzoic acid, Salicylic acid and 4-aminobenzoic acid, and another molecule, anthracene, a rather simple molecule. The performance of predictive (UNIFAC, UNIFAC mod., COSMO-SAC) and semi-predictive (NRTL-SAC) models are compared and discussed according to the functional groups of the molecules and the selected solvents. Moreover, the model errors caused by solid state property uncertainties are taken into account. These errors are indeed not negligible when accurate quantitative predictions want to be performed. It was found that UNIFAC models give the best results and could be an useful method for rapid solubility estimations of an API in various solvents. This model achieves the order of magnitude of the experimental solubility and can predict in which solvents the drug will be very soluble, soluble or not soluble. In addition, predictions obtained with NRTL-SAC model are also in good agreement with the experiments, but in that case the relevance of the results is strongly dependent on the model parameters regressed from solubility data in single and mixed solvents. However, this is a very interesting model for quick estimations like UNIFAC models. Finally, COSMO-SAC needs more developments to increase its accuracy especially when hydrogen bonding is involved. In that case, the predicted solubility is always overestimated from two to three orders of magnitude. Considering the use of the most accurate equilibrium equation involving the ΔC_p term, no benefits were found for drug predictions as the models are still too inaccurate. However, in function of the molecules and their solid thermodynamic properties, the ΔC_p term can be neglected and will not have a great impact on the results.     

Estimating the domain of applicability for machine learning QSAR models: a study on aqueous solubility of drug discovery molecules

We investigate the use of different Machine Learning methods to construct models for aqueous solubility. Models are based on about 4000 compounds, including an in-house set of 632 drug discovery molecules of Bayer Schering Pharma. For each method, we also consider an appropriate method to obtain error bars, in order to estimate the domain of applicability (DOA) for each model. Here, we investigate error bars from a Bayesian model (Gaussian Process (GP)), an ensemble based approach (Random Forest), and approaches based on the Mahalanobis distance to training data (for Support Vector Machine and Ridge Regression models). We evaluate all approaches in terms of their prediction accuracy (in cross-validation, and on an external validation set of 536 molecules) and in how far the individual error bars can faithfully represent the actual prediction error.     

Estimating the domain of applicability for machine learning QSAR models: a study on aqueous solubility of drug discovery molecules

We investigate the use of different Machine Learning methods to construct models for aqueous solubility. Models are based on about 4000 compounds, including an in-house set of 632 drug discovery molecules of Bayer Schering Pharma. For each method, we also consider an appropriate method to obtain error bars, in order to estimate the domain of applicability (DOA) for each model. Here, we investigate error bars from a Bayesian model (Gaussian Process (GP)), an ensemble based approach (Random Forest), and approaches based on the Mahalanobis distance to training data (for Support Vector Machine and Ridge Regression models). We evaluate all approaches in terms of their prediction accuracy (in cross-validation, and on an external validation set of 536 molecules) and in how far the individual error bars can faithfully represent the actual prediction error.     

A knowledge-based approach in designing combinatorial or medicinal chemistry libraries for drug discovery. 1. A qualitative and quantitative characterization of known drug databases

The discovery of various protein/receptor targets from genomic research is expanding rapidly. Along with the automation of organic synthesis and biochemical screening, this is bringing a major change in the whole field of drug discovery research. In the traditional drug discovery process, the industry tests compounds in the thousands. With automated synthesis, the number of compounds to be tested could be in the millions. This two-dimensional expansion will lead to a major demand for resources, unless the chemical libraries are made wisely. The objective of this work is to provide both quantitative and qualitative characterization of known drugs which will help to generate "drug-like" libraries. In this work we analyzed the Comprehensive Medicinal Chemistry (CMC) database and seven different subsets belonging to different classes of drug molecules. These include some central nervous system active drugs and cardiovascular, cancer, inflammation, and infection disease states. A quantitative characterization based on computed physicochemical property profiles such as log P, molar refractivity, molecular weight, and number of atoms as well as a qualitative characterization based on the occurrence of functional groups and important substructures are developed here. For the CMC database, the qualifying range (covering more than 80% of the compounds) of the calculated log P is between -0.4 and 5.6, with an average value of 2.52. For molecular weight, the qualifying range is between 160 and 480, with an average value of 357. For molar refractivity, the qualifying range is between 40 and 130, with an average value of 97. For the total number of atoms, the qualifying range is between 20 and 70, with an average value of 48. Benzene is by far the most abundant substructure in this drug database, slightly more abundant than all the heterocyclic rings combined. Nonaromatic heterocyclic rings are twice as abundant as the aromatic heterocycles. Tertiary aliphatic amines, alcoholic OH and carboxamides are the most abundant functional groups in the drug database. The effective range of physicochemical properties presented here can be used in the design of drug-like combinatorial libraries as well as in developing a more efficient corporate medicinal chemistry library.