Physicochemical QSAR analysis of hERG inhibition revisited: towards a quantitative potency prediction.

In an earlier study (Didziapetris R & Lanevskij K (2016). J Comput Aided Mol Des. 30:1175-1188) we collected a database of publicly available hERG inhibition data for almost 6700 drug-like molecules and built a probabilistic Gradient Boosting classifier with a minimal set of physicochemical descriptors (log P, pKa, molecular size and topology parameters). This approach favored interpretability over statistical performance but still achieved an overall classification accuracy of 75%. In the current follow-up work we expanded the database (provided in Supplementary Information) to almost 9400 molecules and performed temporal validation of the model on a set of novel chemicals from recently published lead optimization projects. Validation results showed almost no performance degradation compared to the original study. Additionally, we rebuilt the model using AFT (Accelerated Failure Time) learning objective in XGBoost, which accepts both quantitative and censored data often reported in protein inhibition studies. The new model achieved a similar level of accuracy of discerning hERG blockers from non-blockers at 10 µM threshold, which can be conceived as close to the performance ceiling for methods aiming to describe only non-specific ligand interactions with hERG. Yet, this model outputs quantitative potency values (IC50) and is not tied to a particular classification cut-off. pIC50 from patch-clamp measurements can be predicted with R2 ≈ 0.4 and MAE < 0.5, which enables ligand ranking according to their expected potency levels. The employed approach can be valuable for quantitative modeling of various ADME and drug safety endpoints with a high prevalence of censored data.

Impact of Established and Emerging Software Tools on the Metabolite Identification Landscape.

Scientists' ability to detect drug-related metabolites at trace concentrations has improved over recent decades. High-resolution instruments enable collection of large amounts of raw experimental data. In fact, the quantity of data produced has become a challenge due to effort required to convert raw data into useful insights. Various cheminformatics tools have been developed to address these metabolite identification challenges. This article describes the current state of these tools. They can be split into two categories: Pre-experimental metabolite generation and post-experimental data analysis. The former can be subdivided into rule-based, machine learning-based, and docking-based approaches. Post-experimental tools help scientists automatically perform chromatographic deconvolution of LC/MS data and identify metabolites. They can use pre-experimental predictions to improve metabolite identification, but they are not limited to these predictions: unexpected metabolites can also be discovered through fractional mass filtering. In addition to a review of available software tools, we present a description of pre-experimental and post-experimental metabolite structure generation using MetaSense. These software tools improve upon manual techniques, increasing scientist productivity and enabling efficient handling of large datasets. However, the trend of increasingly large datasets and highly data-driven workflows requires a more sophisticated informatics transition in metabolite identification labs. Experimental work has traditionally been separated from the information technology tools that handle our data. We argue that these IT tools can help scientists draw connections via data visualizations and preserve and share results via searchable centralized databases. In addition, data marshalling and homogenization techniques enable future data mining and machine learning.

Physicochemical QSAR Analysis of Passive Permeability Across Caco-2 Monolayers.

Caco-2 cell line is frequently used as a simplified in vitro model of intestinal absorption. In this study, a database of 1366 Caco-2 permeability coefficients (Pe) for 768 diverse drugs and drug-like compounds was compiled from public sources. The collected data represent permeation rates measured at varying experimental conditions (pH from 4.0 to 8.0, and stirring rates from 0 to >1000 rpm) that presumably account for passive diffusion across mucosal epithelium. These data were subjected to multistep nonlinear regression analysis using a minimal set of physicochemical descriptors (octanol-water log D, pKa, hydrogen bonding potential, and molecular size). The model was constructed in a mechanistic manner incorporating the following components: (i) a hydrodynamic equation of size- and charge-specific along with nonspecific diffusion across the paracellular pathway; (ii) transcellular diffusion represented by thermodynamic membrane/water partitioning ratio; (iii) stirring-dependent limit of maximum achievable permeability due to the presence of unstirred water layer. The obtained model demonstrates good accuracy of log Pe predictions with a residual mean square error <0.5 log units for all training and validation sets. Given its robust performance and straightforward interpretation in terms of simple physicochemical properties, the proposed model may serve as a valuable tool to guide drug discovery efforts toward readily absorbable compounds.

Compilation and physicochemical classification analysis of a diverse hERG inhibition database.

A large and chemically diverse hERG inhibition data set comprised of 6690 compounds was constructed on the basis of ChEMBL bioactivity database and original publications dealing with experimental determination of hERG activities using patch-clamp and competitive displacement assays. The collected data were converted to binary format at 10 µM activity threshold and subjected to gradient boosting machine classification analysis using a minimal set of physicochemical and topological descriptors. The tested parameters involved lipophilicity (log P), ionization (pK a ), polar surface area, aromaticity, molecular size and flexibility. The employed approach allowed classifying the compounds with an overall 75-80 % accuracy, even though it only accounted for non-specific interactions between hERG and ligand molecules. The observed descriptor-response profiles were consistent with common knowledge about hERG ligand binding site, but also revealed several important quantitative trends, as well as slight inter-assay variability in hERG inhibition data. The results suggest that even weakly basic groups (pK a  < 6) might substantially contribute to hERG inhibition potential, whereas the role of lipophilicity depends on the compound's ionization state, and the influence of log P decreases in the order of bases > zwitterions > neutrals > acids. Given its robust performance and clear physicochemical interpretation, the proposed model may provide valuable information to direct drug discovery efforts towards compounds with reduced risk of hERG-related cardiotoxicity.

Aqueous acidities of primary benzenesulfonamides: Quantum chemical predictions based on density functional theory and SMD.

Aqueous pK(a) of selected primary benzenesulfonamides are predicted in a systematic manner using density functional theory methods and the SMD solvent model together with direct and proton exchange thermodynamic cycles. Some test calculations were also performed using high-level composite CBS-QB3 approach. The direct scheme generally does not yield a satisfactory agreement between calculated and measured acidities due to a severe overestimation of the Gibbs free energy changes of the gas-phase deprotonation reaction by the used exchange-correlation functionals. The relative pK(a) values calculated using proton exchange method compare to experimental data very well in both qualitative and quantitative terms, with a mean absolute error of about 0.4 pK(a) units. To achieve this accuracy, we find it mandatory to perform geometry optimization of the neutral and anionic species in the gas and solution phases separately, because different conformations are stabilized in these two cases. We have attempted to evaluate the effect of the conformer-averaged free energies in the pK(a) predictions, and the general conclusion is that this procedure is highly too costly as compared with the very small improvement we have gained.

Improving the prediction of drug disposition in the brain.

INTRODUCTION: Ability to cross the blood-brain barrier is one of the key ADME characteristics of all drug candidates regardless of their target location in the body. While good brain penetration is essential for CNS drugs, it may lead to serious side effects in case of peripherally-targeted molecules. Despite a high demand of computational methods for estimating brain transport early in drug discovery, achieving good prediction accuracy still remains a challenging task. AREAS COVERED: This article reviews various measures employed to quantify brain delivery and recent advances in QSAR approaches for predicting these properties from the compound's structure. Additionally, the authors discuss the classification models attempting to distinguish between permeable and impermeable chemicals. EXPERT OPINION: Recent research in the field of brain penetration modeling showed an increasing understanding of the processes involved in drug disposition, although most models of brain/plasma partitioning still rely on purely statistical considerations. Preferably, new models should incorporate mechanistic knowledge since it is the prerequisite for guiding drug design efforts in the desired direction. To increase the efficiency of computational tools, a broader view is necessary, involving rate and extent of brain penetration, as well as plasma and brain tissue binding strength, instead of relying on any single property.

QSAR analysis of blood-brain distribution: the influence of plasma and brain tissue binding.

The extent of brain delivery expressed as steady-state brain/blood distribution ratio (log BB) is the most frequently used parameter for characterizing central nervous system exposure of drugs and drug candidates. The aim of the current study was to propose a physicochemical QSAR model for log BB prediction. Model development involved the following steps: (i) A data set consisting of 470 experimental log BB values determined in rodents was compiled and verified to ensure that selected data represented drug disposition governed by passive diffusion across blood-brain barrier. (ii) Available log BB values were corrected for unbound fraction in plasma to separate the influence of drug binding to brain and plasma constituents. (iii) The resulting ratios of total brain to unbound plasma concentrations reflecting brain tissue binding were described by a nonlinear ionization-specific model in terms of octanol/water log P and pK(a). The results of internal and external validation demonstrated good predictive power of the obtained model as both log BB and brain tissue binding strength were predicted with residual mean square error of 0.4 log units. The statistical parameters were similar among training and validation sets, indicating that the model is not likely to be overfitted.

Ionization-specific QSAR models of blood-brain penetration of drugs.

This study presents a mechanistic QSAR (quantitative structure-activity relationship) analysis of blood-brain barrier (BBB) penetration of drugs and drug-like compounds governed by passive diffusion in rats and mice. The analyzed data included a previously compiled set of almost 200 experimental BBB permeation rates (expressed as log PS values) as well as steady-state brain/plasma distribution ratios for ca. 500 compounds (represented as log BB constants) that were considered free of carrier-mediated transport and other unwanted effects. These data were modeled in terms of nonlinear lipophilicity and ionization dependences. The necessity to separate the influence of drug binding to plasma and brain constituents on the distribution ratio is discussed. Preliminary results demonstrate that, if both the rate and extent of BBB penetration are considered, it is possible to estimate whether a given compound may exhibit central nervous system (CNS) penetration.

Ionization-specific analysis of human intestinal absorption.

This study presents a mechanistic QSAR analysis of human intestinal absorption of drugs and drug-like compounds using a data set of 567 %HIA values. Experimental data represent passive diffusion across intestinal membranes, and are considered to be reasonably free of carrier-mediated transport or other unwanted effects. A nonlinear model was developed relating %HIA to physicochemical properties of drugs (lipophilicity, ionization, hydrogen bonding, and molecular size). The model describes ion-specific intestinal permeability of drugs by both transcellular and paracellular routes, and also accounts for unstirred water layer effects. The obtained model was validated on two external data sets consisting of in vivo human jejunal permeability coefficients (P(eff)) and absorption rate constants (K(a)). Validation results demonstrate good predictive power of the model (RMSE = 0.35-0.45 log units for log K(a) and log P(eff)). High prediction accuracy together with clear physicochemical interpretation (log P, pK(a)) makes this model particularly suitable for use in property-based drug design.

Ionization-specific prediction of blood-brain permeability.

This study presents a mechanistic QSAR analysis of passive blood-brain barrier permeability of drugs and drug-like compounds in rats and mice. The experimental data represented in vivo log PS (permeability-surface area product) from in situ perfusion, brain uptake index, and intravenous administration studies. A data set of 280 log PS values was compiled. A subset of 178 compounds was assumed to be driven by passive transport that is free of plasma protein binding and carrier-mediated effects. This subset was described in terms of nonlinear lipophilicity and ionization dependences, that account for multiple kinetic and thermodynamic effects: (i) drug's kinetic diffusion, (ii) ion-specific partitioning between plasma and brain capillary endothelial cell membranes, and (iii) hydrophobic entrapment in phospholipid bilayer. The obtained QSAR model provides both good statistical significance (RMSE < 0.5) and simple physicochemical interpretations (log P and pKa), thus providing a clear route towards property-based design of new CNS drugs.