MM/(GB)(PB)SA approaches in drug discovery are attractive because simplicity and computational speed. For large collections of compounds encountered e.g. in rescoring of high-speed virtual docking results, the choice of the large number of parameters like simulation protocols, force fields, protonation and tautomeric states, various implementations of Poisson-Boltzmann (PB) or Generalized-Born (GB) can be problematic. However, if ranking is more relevant than a close fit to experimental free energies of binding, some choices are less critical. In this work, we concentrated on the protocols to be followed to take into account the variability of molecular-dynamics-based (MD) results. We also checked the possible influence of keeping a small number of selected explicit water molecules as part of the MM/GBSA scheme. Comparisons to simple energy-refined starting structures reveal that for "correct" initial poses, a simple energy refinement followed by an MM/GBSA evaluation can be sufficient (or superior) to lengthy MD treatment.

The selection of an ideal combination of docking algorithm and scoring function is essential for optimal execution of virtual screening campaigns. The performance of docking programs and scoring functions can be evaluated using benchmark data sets for redocking of co-crystallised molecules or active compound enrichment against decoy ligand sets. For drug-like small molecules, several data sets are available. However, for less conventional compounds such as peptides these are missing so far.

In order to provide rational benchmark data sets for these compound classes, the LEADS project (Lessons for Efficacy Assessment of Docking and Scoring) was initiated and a first benchmark data set for peptides (LEADS-PEP) has been released (available on www.leads-x.org). LEADS-PEP consists of 53 protein−peptide complexes that have been selected and prepared using a rational and unbiased workflow. An initial test of several docking programmes and scoring functions revealed large deviations in redocking performance.

The talk will cover the details on the problems of benchmark data sets and provide information on the generation of LEADS-PEP and other data sets as well as their utilization to assess the performance of various popular molecular docking programs and scoring functions.

Computer modeling of protein-protein interactions plays an increasingly important role in studies of biologics. This work presents a method for identifying important interaction sites in protein interfaces and carrying out epitope mapping using the MOE software package. An analysis is carried out of molecular properties mapped onto the protein surface to determine patches which play a role in determining protein properties and binding interactions. A robust first-principles docking calculation is used to generate an ensemble of protein-protein poses which sample the space of relative orientations. An interaction fingerprint encoding the set of contacts between surface patches of different types is used to generate pose clusters which are ranked by ensemble free energy and used to extract the consensus interactions which comprise the predicted epitope for each cluster. This methodology can be combined with experimental data to bias pose generation, cluster ranking, and epitope selection in an integrated process to generate high-quality models. We present a case study in which hydrogen-deuterium exchange data are used to extract the key residue interactions from calculations performed on an ensemble of homology models of the interacting chains.

The understanding of binding interactions between a protein and a small modulator plays a key role in the rationalization of potency and selectivity. However, because of the size of the protein, MM is commonly applied in these large molecules by the use of a “ball-and-spring” model of atoms and molecules with classical forces between them. Conversely, the QM method treats molecules as a collection of nuclei and electrons where geometry and other molecular properties can be obtained by solving the Schrödinger equation. Although the QM models have been shown to be the most accurate they are also the most expensive in terms of time and computational resources, and as a result, are mainly applied on small systems only. We here present a novel QC-based methodology fast enough to be applied in the Drug Design process in a high throughput manner.

Botulinum neurotoxins (BoNTs) provide effective treatment concerning neuromuscular and autonomic conditions. BoNTs are composed of three domains with well-defined structure and function. This flexible modular architecture provides a unique scaffold for the design of therapeutically relevant molecules for a wide range of clinical applications. However, their highly potent nature and resultant low dose, presents some unique analytical challenges. Molecular modelling studies can provide key understanding of structural features and degradation pathways. They can be a powerful tool for rational design of new molecules and support the successful CMC development of therapeutic drugs. In this case study, we present an insight into the quality attributes for toxin therapeutics which were gained with a combined computational and analytical characterization approach.

Protein-display systems, such as phage-display, have demonstrated the power of simultaneously examining: stability; the ability to bind to a target; and the impact of site-specific mutations. Unfortunately, traditional approaches to computer aided molecular design are often intractable and unable to prospectively suggest viable mutations due to the sheer size of the available mutations space. Leveraging the power MOE deployed in a cloud environment, we present a Virtual Phage Display approach which is able to produce predictive models for both thermostability and solubility in under an hour. This MOE-based methodology coupled with the power of the cloud allows modeling to be a prospective exercise, informing and influencing projects from the outset, before time and effort are expended at the bench.

A comparison of drugs with agrochemicals highlights many structural similarities but also clear distinctions. While some molecular feature differences can be rationalized, this is not possible for other cases. The focus in this presentation is on structural elements used in the context of agrochemical research that may have some merit in medicinal chemistry.

Ubiquitin specific proteases (USPs) represent the largest subfamily of deubiquitinating enzymes (DUBs), with more than 50 examples identified within the human genome. USPs are implicated in multiple therapeutic areas, including oncology, immune and inflammatory conditions, and while they are an attractive target class, they have proven challenging to pursue. This presentation will discuss recent developments in the field of small-molecule USP inhibitors, particularly with respect to structural data, and how they may be potentially exploited against the USP target class.

Serine proteases are involved in many diseases and thus over the last decades a very attractive target for the research and development of drugs in pharmaceutical companies. Typically, they modulate regulation or degradation processes in our body, by binding to specific 4 to 8 amino acid residue recognition motifs of other proteins and peptides.

Nature has developed a large repertoire of natural proteins and peptides serving as inhibitors of specific proteases in order to guarantee the proper action of each protease at its dedicated physiological organ and target. This knowledge and available 3D-structures can be used as starting points to establish new protease inhibitors.

Based on the 14-residue cyclic sunflower trypsin inhibitor (SFTI) highly selective inhibitors of serine proteases with a big diversity of cleavage specificities could be designed. The obtained peptidic inhibitors of 3 serine proteases are correlated with computational prediction and design tools available in MOE.

ElectroShape is an established shape-based method for ligand-based virtual screening, which was originally tested on the DUD database and has been used successfully in a number of projects. The method allows ultra-fast comparison of the shapes and charge distributions of molecules in order to rank them by similarity. Consequently, the two most important steps in the preparation of molecules for ElectroShape descriptor generation are the production of suitable conformers and the assignment of atomic charges, both of which have evolved and improved in recent years, as has the benchmark for assessing method performance. We test different conformer generation methods and partial charge calculation schemes, this time using DUD-E database ligand-decoy datasets, to assess the effect of changing these methods on the performance of the ElectroShape approach.

Dr Graeme Robb, Chemistry, Oncology, IMED Biotech Unit, AstraZeneca, Cambridge, UK.

Inhibition of the interaction between B-cell lymphoma 6 (BCL6) and co-repressors has been implicated as a therapeutic target in diffuse large B-cell lymphoma (DLBCL) cancers. Profiling of potent and selective BCL6 inhibitors are critical to test this hypothesis, however inhibition of a protein-protein interaction without prior small-molecule precedent is challenging. We used existing X-ray protein crystal structures to scan the co-repressor sequence for hot-spots and thus identify potential binding-sites for shape-based virtual screening. The results, together with a parallel fragment screen led to the identification of a series of pyrazolo[1,5-a]pyrimidines with weak binding affinity. Early crystallography was critical in enabling structure-based design. The understanding of key affinity interactions within the binding site and the modelling of the dynamic behaviour and free energy of ligand, pocket and solvent facilitated the identification of regions containing unstable solvent molecules, the displacement of which increased affinity 1000-fold. Inhibitor conformations were studied with NMR and we developed a predictive model using mixed Monte Carlo / molecular dynamics simulation, that gave good correlation with observed affinity. We subsequently designed macrocycles to restrict conformational freedom and favour the conformation required for binding, which gave a further 100-fold increase in potency. In an attempt to increase affinity further, we designed several analogues to irreversibly bind to a cysteine residue in the binding site. Despite demonstrating covalent bond formation, these compounds were less effective and we have been able to show, through long time-scale molecular dynamics, that the geometry of the cysteine is poorly available for Michael addition. The use of protein crystal structure, the measurement and prediction of ligand conformation and the use of modern structure-based design techniques has enabled rapid progress from fragment hit to potent BCL6 inhibitor.

Dr Ekaterina Ratkova, Medicinal Chemistry, Cardiovascular, Renal and Metabolism IMED Biotech Unit, AstraZeneca, Gothenburg, Sweden

It is well recognized that interfacial water molecules are of high importance for protein-protein interactions (PPI) and their inhibition. The talk will cover a recent study, where we revealed an extensive network of energetically-favorable water wrapping a newly formed complex between an inhibitor and PEX14 protein. The water-network supported adaptation of low molecular weight inhibitors to a flat, solvent-exposed protein pocket via mediation of polar protein-ligand interactions. We provided a strong evidence of importance of the water-network as an essential binding enhancer in Drug Design. Moreover, we proposed and tested a novel strategy for optimization of PPI inhibitors via extension of the water-network.