Current Computer - Aided Drug Design - Volume 1, Issue 4, 2005

With recent estimates of drug development costs on the order of $800 million and increased pressure to reduce consumer drug costs, it is not surprising that the pharmaceutical industry is keenly interested in reducing the overall expense associated with drug development. An analysis of the reasons for attrition during the drug development process found that over half of all failures can be attributed to problems with human or animal pharmacokinetics and toxicity. Discovering pharmacokinetics and toxicity liabilities late within the drug development process results in wasted resource expenditures. This argues dramatically for evaluation of these properties as early as possible, leading to the concept of "Fail Early". Computational models provide a low cost, flexible evaluation of compound properties that can be implemented and used prior to chemical synthesis thereby creating an alternative philosophy of "Design for Success". Here we review the history and current trends within ADME/Tox modeling and discuss important issues related to development of computational models. In addition, we review some of the commercially available tools to achieve this goal as well as methods developed internally to address these issues from the design stage through development and optimization of drug candidates. In particular, we highlight those features that we feel best exemplify the Design for Success philosophy.

The central nervous system (CNS) is endowed with very efficient protection mechanisms. However, the same mechanisms that protect it, sometimes can be an enemy for therapeutic applications. In this way, many antipsychotic drugs used, are ineffective in the treatment of cerebral diseases such as schizophrenia. Many typical and atypical neuroleptics are very efficient against the positive symptoms, but not against the negative symptoms. To reduce the inefficiency of known neuroleptics, many research efforts have recently focused on the development of new strategies for new neuroleptics drug design. For this reason it was necessary to apply very fast and precise techniques, such as: QSAR (Quantitative Structure-Activity Relationships) and combinatorial chemistry methods, capable to analyze and predict the biological activity for these structures, taking in account the possible changes of the molecular structures. This review intends to detail recent advances in the field of structure-activity relationship and combinatorial chemistry applied to neuroleptics. The antipsychotic activities (log ED50) of potent neuroleptics as indole derivatives, were correlated with pharmacokinetic parameters namely: molecular volume (V), globularity (G), Octanol/water partition coefficient (logP), solubility(S), dipole moment, polarizability. QSAR studies of benzothiazepine derivatives as potent neuroleptics are presented. In addition, the 3D-QSAR methods such as Comparative Molecular Field Analysis (CoMFA) and Comparative Molecular Similarity Indices Analysis (CoMSIA) were applied for a set of dopamine D4 receptor antagonists. The combinatorial chemistry was used to develop a large chemical library starting from 5-hydroxytryptamine2A receptor antagonists used as antipsychotics.

The models available in the literature for the in silico prediction of ADME (absorption, distribution, metabolism, excretion) properties, as well as the data sets used to derive them, are reviewed. Special emphasis is given to describe the latest and most complete models, with the largest and most diverse data sets. Models for human intestinal absorption, oral bioavailability, plasma protein binding, blood-brain barrier permeation, Pglycoprotein substrates and inhibitors, and metabolism are reviewed and discussed. An attempt is made to describe the general picture emerging from each set of models when possible, as well as the issues remaining to address in the different areas for future work. These models are an example of the utility of models and computer simulations for the prediction of pharmacokinetics.

Perhaps the most exciting sub-discipline within Bioinformatics is the application of informatic methods to immunology. Immunoinformatics, which combines informatics with computational chemistry, is facilitating important change within immunology. As it frees itself from the empirical straight jacket that has characterised its development, immunoinformatics is helping immunology to engage fully with the dynamic post-genomic revolution sweeping through bioscience. Focussing on quantitative aspects, we will review recent developments within immunoinformatics, paying particular attention to the following: the development of functional immunological databases; prediction of the antigen presentation pathway; predicting the specificity of peptide Major Histocompatibility Complex (MHC) interactions, using statistical techniques and atomistic molecular dynamics; and the grouping of MHC molecules into supertypes.

Type 1 collagen is the primary protein in extracellular matrix of major tissues. Ligand binding to type 1 collagen is therefore an important problem of interest in areas such as adhesive bonding to tissues, mineralization of collagen scaffolds etc. The triple helical structure of collagen molecule, and the selfassembly of these molecules into fibrils as well as the role of water in its conformational state present interesting challenges in evaluating the binding of ligands to such a structure. Computer simulation of interactions between collagen and other molecular entities (e.g., ligands, proteins, mineral entities etc.) can provide a wealth of information. This paper reviews the computational methods suitable for applications to collagen-ligand binding studies and the current literature on such studies. These methods have been used for indirect (active analog approach) and direct (manual and automatic docking) methods of computer binding simulations. In particular, AutoDock method was extremely valuable to identify the low energy collagenligand complexes, to visualize the hydrogen bonds between collagen and ligands in their complexes, and to characterize the docking/binding energy parameters in the presence of water. Experimental binding assay studies were also used to characterize the interactions. The results give valuable information on criteria for formulation design in practical applications of adhesive bonding to tissues (e.g., bonding of dentin prior to filling of cavities to treat caries). Ongoing current studies also focus on immobilization of charged protein molecules on type 1 collagen to aid in biomimetic mineralization of collagen scaffolds.

Computational ligand docking and screening is widely employed throughout the pharmaceutical industry to speed up the drug discovery process and identify drug candidates from very large pools of virtual compound libraries. When a ligand interacts with a receptor a number of structural changes within the ligand binding site might occur. It is therefore critical for these methods to accurately predict, or otherwise take into account the receptor flexibility upon ligand binding. This flexibility within the binding pocket explains why a diverse range of ligand sizes and shapes can sometimes bind to the same receptor pocket. This observation supersedes the notion that ligand-receptor interaction is a purely 'lock and key' mechanism. The capability to correctly predict molecular interactions is critical for computer-aided molecular design technology. In this review, we discuss biological cases of receptor flexibility upon ligand binding that can range from 'large-scale' movement of loops to single 'gate-keeper' amino acid movements. In addition, we provide further evidence that rigid receptor docking alone will more than likely fail in the drug-discovery process. We then discuss computational methods, which have been developed to mimic flexibility within the binding pocket and predict ligand-receptor interactions. Early flexible receptor docking methods used 'soft-potential docking' and rotamer libraries. More recently methods have focused on constructing an ensemble of structures generated by a variety of means including X-ray crystallography, NMR, Monte Carlo sampling, Normal Modes-based methods and Molecular Dynamics. It is evident that methods that ignore receptor flexibility can result in poorly docked solutions and therefore the challenge is to develop computational methods, which can accurately and efficiently predict this phenomenon.