Hans Matter*, Stefan Güssregen*, Gerhard Hessler*, Stefan M. Kast**
*Sanofi-Aventis Deutschland GmbH, R&D, IDD, Synthetic Molecular Design, Frankfurt am Main, Germany.
** Fakultät für Chemie und Chemische Biologie, TU Dortmund, Dortmund, Germany.

Water molecules are essential to mediate interactions between ligands and their protein binding sites. Understanding the nature and energetics for experimental and postulated hydration sites can significantly impact structure-based design for a project. In particular the displacement of postulated water in favorable hydration sites by ligand moieties with appropriate interactions alters the free energy of binding of a ligand. In this talk computational approaches to characterize hydration sites will be discussed and applied to congeneric series from structure-based projects.

3D RISM (3D reference interaction site model) calculations as prominent approach to understand thermodynamic features of hydration sites constitute an implicit solvent model based on classical density functional theory, which provides equilibrium thermodynamic quantities. The results are equivalent to those of Monte Carlo or Molecular Dynamics simulations at a fraction of the computational cost.

The application of 3D-RISM fields to fragment deconstruction for protease inhibitors will be discussed. Moreover, ligands with X-ray structures will be investigated and a qualitative agreement to SAR trends is shown. It will be also shown that machine learning employing 3D RISM derived solvation fields can extract SAR trends for on a more quantitative basis to provide further guidance for lead optimization. This led to develop relevant 3D-QSAR models for affinity estimation in congeneric series.

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

Over the last decade, the analysis of water molecules solvating protein-ligand binding surfaces was demonstrated to be highly valuable for structure-based drug design. Potency of inhibitors could be modulated by orders of magnitude using such strategies as displacement of energetically-unfavorable water molecules, functional replacement of energetically-favorable waters or their stabilization as bridges between ligand and protein, and recently discovered optimization of the water network wrapping a newly formed complex surface.

3D RISM approximation of the Integral Equation Theory is an efficient technique of the hydration site analysis, which was implemented in MOE software in a very user-friendly manner. Basics of the theory will be given to provide better understanding of the Solvent Analysis setup and analysis of the results.

Some tips will be given to improve prediction of hydration sites position for the following cases: (i) isolated water molecules in a buried lipophilic cavity of enzymatic pockets, and (ii) water-bridges at solvent-exposed part of the pockets, as well as (iii) a network of water molecules in a cavity of bromodomain proteins, and (iv) an envelope of water molecules around an inhibitor of protein-protein interactions.

Moreover, analysis of solvation shells of interacting partners before binding event could be of value to judge predicted thermodynamic profile of hydration sites and guide decisions on ligand modifications.

Within mainstream technology companies, the virtual assistant concept has been heavily adopted, despite these companies having existing well-developed user interfaces. These include the likes of Google Assistant, Apple’s Siri, Microsoft’s Cortana, Amazon’s Alexa and Samsung’s Bixby, each providing these companies different ways of interacting with users. One of the keys of the virtual assistant concept is providing the user with the information they want at the exact time that they want it. An example would be the way Google delivers information on journey time and traffic to android phones just before the time people may begin their commute to and from their place of work.

Typically, data analysis or the application of computational approaches occurs at the request of users (possibly run by the user themselves). With the virtual assistant concept, this can happen in a far more proactive and automated manner. Responding to the generation of new data, and automatically running predefined analyses. This brings standardization, ease of application across projects, and delivery of information to users they may not have been exposed to otherwise.

Here, we introduce Kernel Eli Lilly, a virtual assistant created to enhance medicinal chemistry at Lilly, along with the wider virtual assistant concept as applicable to medicinal chemistry. Kernel interacts with users (medicinal chemistry teams), delivering automated analysis and results of computational work directly to their email inbox. A number of implemented POC scenarios will be described including a detailed summary and assessment of primary biochemical assay data, as well as de novo design on newly tested compounds, which are then ranked and then sent back to the original submitter.

The talk will conclude with a description of the first example of a fully automated medicinal chemistry design cycle at Lilly. Kernel identified newly tested molecules, designed derivatives, ranked them using predictive models, identified synthetic routes, and then a number were made via automated synthesis. Achieving this in the absence of human intellectual input was a landmark moment bringing together numerous technologies at Lilly and highlighting how automated approaches can complement human scientists.

Dr. Martin Packer, Principal Scientist, Computational Chemistry, Oncology, AstraZeneca, Alderley Park, UK

NMR data often reveals solution-state conformational preferences for drug-like molecules, whether through chemical shifts, coupling constants, NOE intensities or residual dipolar couplings. We consider here information that is available, “in plain sight”, in 1D proton spectra. The use of such spectra to infer conformational preferences is more efficient and effective than use of any other NMR parameter: data are routinely available for all molecules synthesised during design and chemical shifts can be confidently predicted across large conformational ensembles. We review 1D NMR signatures that emerge from closely related acyclic and macrocyclic compounds, and explore the extent to which conformational variation of proton shift can be used to understand solution-state ensembles. We also discuss spectral analysis tools implemented in MOE to facilitate this analysis.

John Porter^a, Tom Coulter^a, Richard Perrins^a, Yao Ding^a, Michael P Mazanetz^b
a) MIDATECH LTD, 65 Oddfellows House, 19 Newport Road, Cardiff CF24 0AA
b) NovaData Solutions Ltd., PO Box 639, Abingdon-on-Thames, Oxfordshire, OX14 9JD, United Kingdom

Gold nanoparticles (GNPs) have shown promise as nanomedicines as they are non-toxic, easy to prepare with controlled size distributions and functionalisable with a range of ligands. The ability to construct tuneable, multi-modal GNP entities allows the precise control of surface properties for targeting, stability and, importantly, the release of therapeutic payloads and GNPs have been reported as potential cancer therapies.

Maytansine, an ansa macrolide, is a highly potent antimitotic agent that exerts an antiproliferative effect by inhibiting microtubule assembly by binding to tubulin with a K_D of around 1 μmol/L. Despite a promising in vitro profile, clinical trials with maytansine in cancer patients failed because of poor efficacy and unacceptable systemic toxicity.

Maytansinoid analogues were prepared and appeared to be an obvious candidate for conjugation to a GNP based on its relative ease of access, ability to conjugate to gold, cytotoxic potency and clinical validation.

A 3D in silico protein model based on the human beta-2-tubulin protein sequence was built using MOE from report X-ray structures in the literature. A schema was developed to generate a conformational ensemble of macrocycles for docking studies using KNIMOE. This scheme was used to prepare a data set ligands for a docking study using MOE docking.

Analysis of a number of significant protein-ligand interactions were identified using the protein-ligand interaction fingerprint (PLIF) tool from MOE. Further analysis of ligand binding using 3D-RISM could rationalise ligand potency and has paved the way forward for the design on the next generation of Maytansinoid analogues.

Inhibition of ATM and Rad3-related (ATR) kinase provides a unique opportunity to disable the DNA damage response as a resistance mechanism in cancer with minimal toxicity to normal cells. In order to optimize an amino pyrazine HTS hit, we constructed homology models of ATR and the related kinases DNA-PK and ATM based on PI3K-, the closest related kinase (~22% ID) for which crystal structures were available. These models guided the introduction of key substituents to modulate selectivity against related kinases involved in the DNA damage response. Inhibitory potency was further fine-tuned by replacing an amide linker by aromatic heterocycles that were predicted to favor the proposed bio-active conformation by means of quantum chemical calculations. Potent activity in cell-based assays and aqueous solubility was achieved by introducing amines in a uniquely negatively charged area of the ATR active site. This combination of incremental improvements, guided by computational methods, resulted in the discovery of M6620/VX-970, a potent and selective ATR kinase inhibitor currently in Phase I/II clinical studies for the treatment of solid tumors.

Aminoacyl tRNA synthetases (aaRSs) form a key link in the protein synthesis process, facilitating the accurate attachment of the appropriate amino acid to the tRNA prior, to transfer to the ribosome where it is added to the growing protein chain. There is significant difference between prokaryotic and eukaryotic synthetases allowing for the use of aaRS inhibitors and antibacterials. aaRSs form a validated class of drug targets, with the natural product mupirocin being an IleRS inhibitor and a topical treatment for Gram-positive infections.

We have developed inhibitors of LeuRS that target the synthetic catalytic site, thus bypassing the editing site that is so prone to resistance mechanisms, as targeted by compounds such as GSK-052. The compounds solely target Gram-negative organisms and have shown activity against a variety of the critical ESKAPE pathogens. The compounds show a Frequency of Resistance that is an order-of-magnitude better than GSK-052 and the resistant mutants generated have shown no mutations in the coding portion of the synthetase gene.