Current Computer - Aided Drug Design - Volume 2, Issue 1, 2006

The huge data sets produced by high-throughput screening (HTS) technologies have created a tremendous challenge for the drug discovery industry. Rapid processing of HTS data and identification of hits are essential in order to accelerate the discovery of quality lead compounds. In addition to finding active compounds among those screened, it is useful to identify the molecular features associated with the activity. To do this, one needs to analyze the initial HTS data to find quantitative relationships between biological activity and specific compound features. There are several challenges in the development of biological activity models from HTS data. First, the hit compounds belonging to different chemotypes may be acting via different mechanisms. Second, many HTS data sets have substantial measurement errors. Third, despite of large exploratory sets which may include thousands of compounds, HTS programs usually provide relatively few active compounds. Powerful and flexible data management systems are key to addressing these challenges. In this review, we elucidate the modern approaches to processing HTS data and developing biological activity models. In our opinion, such systems provide a functional interface between real and virtual screening programs. The synergy of these powerful technologies will increase the efficiency with which high quality clinical candidates are produced, thus providing a great benefit to the industry.

Non-nucleoside reverse transcriptase inhibitors (NNRTIs) are promising compounds in the search for potent and selective drugs for the treatment of AIDS. Although they are structurally diverse, the NNRTIs exhibit a striking similarity in their mode of action on the reverse transcriptase (RT) hydrophobic pocket. Several quantitative structure-activity relationship (QSAR) studies have been devoted to the HEPT and TIBO derivatives acting as NNRTIs. Thanks to their ability to perform non-linear mapping of the physicochemical descriptors to the corresponding biological activity, neural networks (NNs) proved to be a powerful QSAR modeling technique for this series of inhibitors. They show the importance of hydrophobic character of these compounds in their anti-HIV activity variation. One of the purposes of QSAR analyses is to understand the forces governing the activity of a particular class of compounds and to assist drug design. The present work rationalizes in depth the relationship between the hydrophobic character of NNRTIs and their anti-HIV activity. The variation of anti-HIV activity with respect to the hydrophobic parameters is performed by means of NNs and its non-linear aspect is discussed. There is a similarity in the hydrophobic character of TIBO and HEPT derivatives.

This review focuses on both physicochemical and theoretical QSAR methods for the prediction of drug transport across the blood-brain barrier (BBB). Special emphasis is given to the recent progress that has been made in the modeling of BBB penetration, with a particular focus on the models based on kinetic parameters of BBB permeability dataset. Physicochemical models based on partition coefficients and chromatographic capacity factors, as well as computerized parameters such as polar surface area and hydrogenbonding descriptors are described and their success and limitations are discussed. Theoretical models based on topological or molecular orbital calculations are summarized and assessed in terms of descriptors, model type, predictive performance and interpretability. Strengths and weaknesses of the various methods are described. Related issues that are mentioned include the transporter-mediated permeation of drugs across the BBB and its implications on the stability and predictive quality of QSAR models.

This work is concerned with rational drug design at the atomic level. Some fundamental stages of rational drug design are addressed, namely the atomic level understanding on disease-related enzymatic mechanisms and inhibition, which is a pre-requisite to any attempt to rationally design new, better inhibitors; the rational optimization of a lead compound, through chemical modifications that increase its affinity and/or specificity for the target receptor; and the design of new drugs based mainly on the knowledge of the electrostatic potentials of the drug and its receptor, with simultaneous optimization of shape and size complementarity. All methods described here offer interesting approaches to rational drug design, the choice of the method being dependent on the amount of previous knowledge of the system. Examples include the study of the inhibition mechanism of Class Ia Ribonucleotide Reductase, an important anti-cancer target, a brief description of the optimization of the anti HIV lead compound 15-Deoxy-Δ12,14-Prostaglandin J2, through chemical substitutions in the original drug, which is an antagonist for the Nuclear Factor Kappa B:DNA binding, thus precluding HIV gene expression, and the development of new anti-malaria drugs, mainly based on the shape/size of the receptor complemented with the analysis of the electrostatic potentials. A large number of computational techniques are needed for these approaches, ranging from the high level quantum mechanics to the more approximate docking calculations, with the intermediate level hybrid methods and molecular mechanics included. The basic principles for the applications of these techniques are also discussed.

In this review we describe progress in docking and especially high-throughput docking (HTD) for applications in drug design and in silico screening. Computational methods that are used in HTD to assist drug design involve two steps: docking and scoring. Several current docking programs have the ability to generate protein-ligand configurations that are close to the correct structure, as revealed by X-ray crystallography, in many cases. Recent comparison studies of docking and scoring methods have shown that the choice of the best docking (and scoring) tool is to a large extent target-dependent. Most of the docking programs treat the ligand as flexible, but the protein conformation is kept rigid. We review algorithmic advances that allow for partial treatment of protein flexibility. The estimation of binding affinities (scoring) is, from a theoretical point of view, the most challenging part of ligand design. Despite significant progress, a fast and accurate computational prediction of binding affinities is still beyond the limits of current methods. We discuss multivariate statistical methods that have been proposed recently for improving HTD results and briefly outline simplified free energy calculations based on molecular dynamics simulations. Recent applications, in particular those based on end-point free energy models, are re-establishing the interest in this approach and we present an example from our research work. Finally, we discuss new application scenarios of HTD, including chemogenomics docking on entire protein families and docking of natural products.