M. Paul Gleeson

Generation of a Set of Simple, Interpretable ADMET Rules of Thumb

A set of simple, consistent structure-property guides have been determined from an analysis of a number of key ADMET assays run within GSK: solubility, permeability, bioavailability, volume of distribution, plasma protein binding, CNS penetration, brain tissue binding, P-gp efflux, hERG inhibition, and cytochrome P450 1A2/2C9/2C19/2D6/3A4 inhibition. The rules have been formulated using molecular properties that chemists intuitively know how to alter in a molecule, namely, molecular weight, logP, and ionization state. The rules supplement the more predictive black-box models available to us by clearly illustrating the key underlying trends, which are in line with reports in the literature. It is clear from the analyses reported herein that almost all ADMET parameters deteriorate with either increasing molecular weight, logP, or both, with ionization state playing either a beneficial or detrimental affect depending on the parameter in question. This study re-emphasizes the need to focus on a lower molecular weight and logP area of physicochemical property space to obtain improved ADMET parameters.

The importance of the domain of applicability in QSAR modeling

The domain of applicability is an important concept in quantitative structure activity relationships (QSAR) that allows one to estimate the uncertainty in the prediction of a particular molecule based on how similar it is to the compounds used to build the model. In this paper we discuss this important concept, providing details of the development and application of a method to compute the domain of applicability within model descriptor space and structural space as defined by daylight fingerprints. The importance of the domain of applicability is illustrated using five QSAR models generated on plasma protein binding and P450 inhibition datasets. Such methodologies will be shown to offer us a means to monitor the performance of QSARs over time, providing us both with a way to estimate the accuracy of a given prediction and identify when a model needs to be rebuilt.

In-silico ADME models: a general assessment of their utility in drug discovery applications.

ADME prediction is an extremely challenging area as many of the properties we try to predict are a result of multiple physiological processes. In this review we consider how in-silico predictions of ADME processes can be used to help bias medicinal chemistry into more ideal areas of property space, minimizing the number of compounds needed to be synthesized to obtain the required biochemical/physico-chemical profile. While such models are not sufficiently accurate to act as a replacement for in-vivo or in-vitro methods, in-silico methods nevertheless can help us to understand the underlying physico-chemical dependencies of the different ADME properties, and thus can give us inspiration on how to optimize them. Many global in-silico ADME models (i.e generated on large, diverse datasets) have been reported in the literature. In this paper we selectively review representatives from each distinct class and discuss their relative utility in drug discovery. For each ADME parameter, we limit our discussion to the most recent, most predictive or most insightful examples in the literature to highlight the current state of the art. In each case we briefly summarize the different types of models available for each parameter (i.e simple rules, physico-chemical and 3D based QSAR predictions), their overall accuracy and the underlying SAR. We also discuss the utility of the models as related to lead generation and optimization phases of discovery research.

Generation of in-silico cytochrome P450 1A2, 2C9, 2C19, 2D6, and 3A4 inhibition QSAR models

In-silico models were generated to predict the extent of inhibition of cytochrome P450 isoenzymes using a set of relatively interpretable descriptors in conjunction with partial least squares (PLS) and regression trees (RT). The former was chosen due to the conservative nature of the resultant models built and the latter to more effectively account for any non-linearity between dependent and independent variables. All models are statistically significant and agree with the known SAR and they could be used as a guide to P450 liability through a classification based on the continuous pIC50 prediction given by the model. A compound is classified as having either a high or low P450 liability if the predicted pIC50 is at least one root mean square error (RMSE) from the high/low pIC50 cut-off of 5. If predicted within an RMSE of the cut-off we cannot be confident a compound will be experimentally low or high so an indeterminate classification is given. Hybrid models using bulk descriptors and fragmental descriptors do significantly better in modeling CYP450 inhibition, than bulk property QSAR descriptors alone.

QM/MM As a Tool in Fragment Based Drug Discovery. A Cross-Docking, Rescoring Study of Kinase Inhibitors

The use of QM/MM based methods to optimize and rescore GOLD derived cross-docked protein-ligand poses has been investigated using a range of fragment-like kinase inhibitors where experimental data have been reported. Particular emphasis has been placed on rationalizing the potential benefits of the method in the increasingly popular fragment based drug discovery area. The results of this cross-docking, rescoring study on 9 protein ligand complexes suggest that the hybrid QM/MM calculations could prove useful in kinase fragment based drug discovery (FBDD). B3LYP/6-31G**//UFF derived enthalphies allow us to identify the correct X-ray pose from a range of plausible decoys 77% of the time, almost a doubling of the retrieval rate compared to GOLD (44%). In addition, this method provides us with a means to rapidly and accurately generate virtual protein-ligand complexes that will allow a program team to probe the existing interactions between the ligand and protein and search for additional interactions.

QM/MM Calculations in Drug Discovery: A Useful Method for Studying Binding Phenomena?

Herein we investigate whether QM/MM could prove useful as a tool to study the often subtle binding phenomena found within pharmaceutical drug discovery programs. The goal of this investigation is to determine whether it is possible to employ high level QM/MM calculations to answer specific questions around a binding event in a cycle time that is aligned with medicinal chemistry synthesis. To this end QM/MM calculations have been performed on four protein kinase-ligand complexes using five different levels of theory, using standard hardware, in an effort to assess their utility. We conclude that the accuracy and turnaround time of such calculations mean they could prove valuable to (1) probe the subtle nature of the interactions within protein active sites, (2) facilitate the interpretation of poorly resolved electron density, and (3) study the impact of substituent changes on the binding conformation or in the assessment of alternate scaffolds. In practice, the successful application of such methods will be limited by the size of the system under investigation, the level of theory used, and whether there is a need for conformational sampling.

Influenza A virus molecularly imprinted polymers and their application in virus sub-type classification

In this work, we apply a molecular imprinting strategy as a screening protocol for different influenza A subtypes, namely H5N1, H5N3, H1N1, H1N3 and H6N1. Molecularly imprinted polymers for each of these subtypes lead to appreciable sensor characteristics on a quartz crystal microbalance leading to detection limits as low as 105 particles per ml. Selectivity studies indicate that each virus is preferably incorporated by its own MIP. Recognition in most cases is dominated by the neuraminidase residue rather than the hemagglutinin. Multivariate analysis shows that the sensor responses can be correlated with the differences in hemagglutinin and neuraminidase patterns from databases. This allows for virus subtype characterization and thus rapid screening.

Assessing the drug-likeness of lamellarins, a marine-derived natural product class with diverse oncological activities.

Natural products currently represent an underutilized source of leads for the pharmaceutical industry, especially when one considers that almost 50% of all drugs were either derived from such sources or are very closely related. Lamellarins are a class of natural products with diverse biological activities and have entered into preclinical development for the treatment of multidrug-resistant tumors. Although these compounds demonstrated good cell penetration, as observed by their low microM activity in whole cell models, they have not been extensively profiled from a physicochemical point of view, and this is the goal of this study. For this study, we have determined the experimental logP values of a set of 25 lamellarins, given it is the single most important parameter in determining multiple ADMET parameters. We also discuss the relationship between this natural product class, natural product derivatives in development and on the market, oral marketed drugs, as well as drug molecules in development, using a range of physicochemical parameters in conjunction with principal components analysis (PCA). The impact of this systematic analysis on our ongoing medicinal chemistry strategy is also discussed.

A novel approach to identify molecular binding to the influenza virus H5N1: screening using molecularly imprinted polymers (MIPs)

In this report we investigate whether a molecularly imprinted polymer (MIP) of an inactivated strain of influenza A H5N1 could be used to help identify molecules capable of binding to, and inhibiting the function of the virus, via either competitive or allosteric mechanisms. Molecules which bind to the virus and induce a conformational change are expected to show reduced binding to the H5N1 specific MIP. Given the importance of molecular recognition in virus replication, such conformational change might also reduce the effectiveness of neuraminidase (N1) for cleaving the sialic groups necessary for virus replication. We show that the method can indeed differentiate between a potent neuraminidase inhibitor, H1 and H5 antibodies, and N1 specific and non-specific monosaccharide substrates. We suggest that such a method could potentially be used in conjunction with traditional biochemical assays to facilitate the identification of molecules functioning via novel modes of action.

Elucidating the origin of the esterase activity of human serum albumin using QM/MM calculations.

Human serum albumin (HSA) is a critical transport plasma protein accounting for ∼60% of the total protein content in blood. Remarkably, this protein is also found to display esterase activity. In this study, we apply theoretical studies to elucidate the origin of the esterase-like activity arising from the Sudlow site I. Using MD and QM/MM calculations, we investigate which active site residues are involved in the reaction, and the precise mechanistic sequence of events. Our results suggest Lys199, His242, and Arg257 help give rise to the esterase activity and that the most catalytically efficient active site configuration requires that both Lys199 and Aspirin are in their neutral forms. The abundance of HSA in the body suggests the protein might be a suitable target for the computational guided design of acetyl based pro-drugs of acidic molecules that often displayed limited oral exposure due to their unmasked ionizable substituent.