In Silico Lead Discovery

Full text available

The number of new drug approvals per year has been decreasing over the past decade for numerous reasons including an increase in regulatory requirements and lack of sufficient knowledge on the pathophysiological processes being targeted. Also, many compounds fail in development because they lack efficacy and safety. One strategy, among many, to circumvent this high attrition rate is through improving the quality of the compound collections as libraries have usually grown in size with little or inadequate attention about the quality. There are many different ways to prepare a compound collection, the process can involve increasing diversity but it can also imply the creation of focused collection dedicated to a specific disease-type and/or target. In parallel, ADMET (Absorption, Distribution, Metabolism, Excretion and Toxicity) properties have to be considered and the parameters assessed tuned according to the project, stage of the project and disease type. In this chapter, in silico methods facilitating the creation of a generic target-independent compound collections are explored and several in silico ADMET prediction tools are discussed. Key concepts are described to build a compound collection appropriate for hit finding. Overall, this procedure involves several steps: database cleaning, compound filtering step using drug-likeness or lead-likeness criteria, removal of undesirables chemical structures, and structural and chemical quality control. Because for in silico screening studies the collection has to be in 3D, a paragraph exposes some recently developed methods for 3D structure generations, and a list of commercial and free standalone packages as well as online tools is provided.

The number of promising macromolecular targets involved in drug discovery programs has considerably increased because of the recent advances in human genomics and proteomics. In this respect, modern techniques like virtual screening in combination with high-throughput screening are well established approaches to assist identification of novel lead compounds. In particular, structure-based virtual screening is widely applied to search for new hit molecules among a large number of chemical compounds against therapeutically relevant protein targets with known three-dimensional structures. Here, we introduce the main techniques and programs applied for structure-based virtual screening with a focus on the molecular docking–scoring methodology and we discuss key issues that still need to be improved. In addition, we give recent examples of successful applications of hierarchical structure-based virtual screening methods combined with ligand-based ones.

The search for new lead compounds when no sufficient structural information on the receptor binding site, neither on the binding mechanism, are available is challenging, but it can be proceeded if biological activity data are provided for at least one active compound towards the target under consideration. In the recent years, traditional approaches with respect to this situation such as QSAR and pharmacophore techniques have been supplemented by 3D similarity search techniques to mine large banks of compounds, searching for small compounds similar to the bioactive ones. We introduce some of the concepts related to the latter techniques, and we discuss briefly foreseen developments.

Fragment-based drug discovery (FBDD) has received much attention in recent years by the pharmaceutical industry and academia. Here published FBDD methods are reviewed with a particular focus on in silico methods and theoretical approaches. Included are techniques useful at each stage of the process, from fragment library design, to virtual fragment screening and on to fragment growth and lead generation. The strengths and weaknesses of FBDD approaches are presented and the benefits of the complementary use of experimental and computational methods are discussed.

This chapter outlines the progress made in analyzing and predicting binding pockets. Typical structurebased drug discovery project begins with 3D structure of the target protein, identifies putative binding pocket(s), analyses its properties and performs virtual screening to find suitable candidate drug molecules. The success crucially depends on both the correct binding pocket prediction and revealing essential biophysical characteristics of the binding site. These two tasks are intertwined, since many of the binding pocket prediction methods relay of previous studies on binding site physico-chemical properties. In this chapter the most popular methods and approaches for both analysis and prediction of binding sites are reviewed and the corresponding URLs are provided. The emerging picture is the most successful methods of binding site prediction are based on many components analysis and thus reflect the complex nature of the receptor-ligand interactions.

It is now well recognized that receptor flexibility plays an important role in protein-ligand binding. This flexibility can concern not only the mobility and reorganization of side chains in the binding site but also conformational changes of the whole molecule, which makes modeling of ligand docking more challenging. We give here an overview of existing approaches to treating receptor flexibility in protein-ligand docking and virtual screening that range from approximate to more accurate methods, including the use of normal modes in accounting for global conformational changes of the receptor or the use of more precise force fields. In addition, we describe recent successful applications of such approaches that have led to the design or discovery of new lead compounds with therapeutic relevance. These new developments in protein-ligand docking and screening are increasingly applied for better prediction of binding affinities. In view of the ever-increasing power of machine computation, a better accounting of the flexibility of the receptor is justified in order to improve the prediction of ligand binding in the search for new drug candidates by virtual screening.

Protein–Protein Interactions (PPIs) constitute a promising class of targets for drug discovery. Inhibitors of these original interactions are certainly the next generation of highly innovative drugs that will reach the market in the next decade. However, the in silico design of such compounds still remains challenging. This review describes this particular protein-protein interaction chemical space and the main biophysical reasons that make them challenging targets for the drug discovery process. A state of the art of protein databases and servers dedicated to PPIs, their analysis and inhibition is also surveyed. It then presents some innovative methodologies that led to the development of new inhibitors. Finally, different families of protein-protein interactions for which an inhibitor is known are briefly introduced and potential tracks for the future are proposed such as a new classification based on protein-protein interfaces with known inhibitors into specific families, with a subsequent notion of focused databases dedicated to each specific class.

ABC transporters are involved in variety of processes of physiological and clinical significance. Besides their function as natural physiological protectors of the living organisms against xenobiotics, they play a crucial role in drug pharmacokinetics and for the multidrug resistance in tumor cells. The in silico modeling of ABC transporters is extensively developing in the recent years making use of increasing data about 3D structures of transport proteins. The chapter describes the most recent achievements in the computational studies starting from ligand-based design approaches and classification algorithms to homology modeling and docking of ligands. Many of the models show a satisfying performance shading light on the structure-function relationships of the proteins and their substrates and inhibitors, as well as generating hypotheses of ligand-protein interactions and helping in design of further experimental studies. However, a number of problems related to reliability of the experimental data used for modeling and computational methodologies applied limit the applicability of these methods for virtual ligand screening. First attempts towards docking of ligands into transporter binding sites are discussed to illustrate these limitations.

Rational drug design approaches represent an important alternative strategy to the discovery of new therapeutics based on simply serendipity. Computational chemists have to understand the peculiar aspects linked to the biological system in study and on this basis choose the right collection of complementary in silico approaches to tackle in the better way the medicinal chemistry problem in study. In this chapter several examples of successful applications of in silico approaches for lead/drug discovery are presented. In the first paragraph are discussed interesting structure based methods to design inhibitors of protein kinases, while in the second both structure and ligand based techniques are exploited in a consensus fashion to develop novel Adenosine A3 receptor antagonists. The last paragraph presents an overview of ligand based virtual screening and an example of a successful application of this approach to identify a small molecule agonist of the NTS1 receptor. Even if the cases reported cannot cover completely the complexity of the computational techniques available nowadays, they embody different interesting examples useful to understand the potentialities, limits and pitfalls of modern molecular modeling approaches.

A key aspect of efficient scientific research is the efficient accumulation, exchange and presentation of relevant information. Multi-disciplinary fields of research, such as structure-based lead discovery (SBLD), depend upon many different data types and require the concurrent capture and display of such heterogeneous types of data. The visualisation of such data in a more intuitive and accessible format than the underlying data capture method is often extended to provide the viewer context within a wider range of data types. In the field of SBLD, experimental protein structure determination presents a need to visualise three-dimensional data and to further annotate such visualisations with additional information. The use of high-throughput methods in both chemistry and biology has resulted in a rapid accumulation of relevant information to support and prioritise SBLD. The appropriate integration and visualisation of this data maximises the impact of the underlying information within the context of a project both not only for computational chemists and structural biologists but also for biologists and medicinal chemists. In this chapter we will discuss current approaches and outstanding issues associated with this particular challenge in the context of SBLD.

Full text available