Chimeric protein degraders (ProDegs, PROTACs) exert their efficacy by simultaneously binding an E3 ligase and a protein target of interest. This necessary first step results in protein target poly-ubiquitination and subsequent proteasomemediated degradation. Compound optimization typically involves identifying a target binder and an E3 ligase binder and connecting the two pieces by varying linkers. Potent and selective degradation of various proteins have been reported in the literature. Recent structural studies of ternary complexes attempt to provide a detailed understanding of the molecular mechanism of action. This presentation will introduce this novel modality, identify challenges and opportunities for computational modeling in this space with a specific study on BTK.

Recent step changes in the availability of cheap and dense compute power have enabled the routine scale-up of “one-off” descriptor calculations from handfuls at a time to tens of thousands or more. This complements the current explosion in high throughput experimental data collection and analytics techniques. Indeed, when coupled together, high-throughput computation and experimentation can be used to inform this generation of machine learning approaches, pragmatically guide small molecule development, or rapidly assess the tractability of protein targets. Of particular interest to GSK and the broader drug discovery community is the identification of protein cavities that have the potential to bind small molecules. Owing to the importance of this task, numerous cavity detection methods exist employing both static and dynamic representations of the protein system. In this talk, I will discuss the design and implementation of a high throughput computational protocol that employs these various detection methods, as well the challenges associated with scaling the protocol to identify and assess the makeup of cavities present throughout the human proteome.

Molecular visualization with stereoscopic graphics systems has been at the core of structure-based drug design since its inception, because stereo vision on a monitor or projector can enable better understanding of the complex 3D data found in binding sites. The base technology to enable this visualization, however, has not progressed over decades of use. Some niche solutions and improvements have come and gone. Indeed, with the departure of passive and autostereoscopic monitors (monitors that do not require special graphics cards or powered glasses), there is significant risk of regression. In this talk, we will review the role of stereoscopic visualization in drug discovery, discuss what we think are opportunities that are missing in our ability to analyze 3-dimensional structural information, and describe the porting of MOE to a Zvr virtual reality monitor that enables a more enhanced 3D visualization experience.

The talk will show the basic methods to obtain descriptors of protein surfaces which are of general use in AI powered classification methods. We propose that the inherent restrictions on the nature of epitope sequences as well as the discreet architecture usually present in the paratopes provide a fair ground to develop surface descriptors focused on these areas. These epitope-paratope descriptors can be used for prediction of affinity and stability of therapeutic antibodies. We believe that the need for accurate descriptors of protein surfaces is leading the structural analysis to focus on new structural features and patterns and pushing this discipline into a new arena.

Jeremy DESAPHY, Cen GAO, Jibo WANG
During the drug discovery phase, many crystal structures of a target in complex with several ligands are solved. Depending on the target, its flexibility, the binding site complexity, it becomes rapidly difficult for one to get a clear and complete picture. When structure-based design projects arise, articulation of molecular hypotheses become more challenging, more precise, and often requires a manual and time-consuming analysis. With the additional layer of inconsistencies on the protein side – insertion / deletion / mutation / inconsistent numbering – it is necessary to create a new system that addresses such issues while opening the door to fast querying capabilities.

Over the last few years, technology has enabled a wider range of opportunities, leading to improve performance and ease to use. In this presentation, we will present our strategy to provide a wider range of querying capabilities to both medicinal chemists and computational chemists. This new system automatizes some daunting tasks that computational chemist is required to do while giving them the freedom to ask more complex questions. At last, we will present some applied examples of our work and future directions.

I’ll present three stories on approaches for tough targets. The first is a computational method for rapidly identifying cryptic pockets that have high potential to be druggable. The second is around drug design approaches contributing to the discovery of a protein-protein inhibitor. The third is around design of a hybrid antibody-small molecule tool for cell-surface targets difficult to modulate by either large molecule or small molecule alone.

One potential approach for using molecular dynamics (MD) trajectories of proteins or protein–small molecule complexes as starting points for virtual screening and de novo design relies on the use of pharmacophores. In the case of a single structure, pharmacophore generation is usually performed manually, which is time-consuming and subjective, and therefore not easily applied to the many structures coming from an MD trajectory. Here we describe a fast and automated procedure that we have developed to generate pharmacophores from either holo- or apo-protein structures. We demonstrate the usefulness of such automatically generated pharmacophores using the DUD-E virtual screening validation set in combination with our in-house virtual screening protocol. We found that our pharmacophore-based virtual screening approach not only greatly outperforms other pharmacophore-based methods for which results are available, but that it also shows performance comparable to the best-available docking methods (judged according to standard metrics). We also attained a significantly higher early enrichment compared to docking, which should provide a practical advantage in drug discovery applications.

Cytochrome P450 oxidases (CYPs) are a class of well-known heme-containing enzymes that are primarily responsible for clearing xenobiotics through oxidative metabolism. In humans, the xenobiotics targeted by CYPs include small molecule therapeutics, and thus understanding the interplay between drug molecules and CYPs is critical for evaluating drug efficacy, clearance, toxicity, and drug-drug interactions. About 80% of CYP activity in humans is dominated by five CYP isoforms, which differ in sequence and somewhat in structure (particularly near the binding pocket around the heme). Although dozens of crystal structures of the five primary CYP isoforms have been solved, most modeling tools to predict drug-CYP interactions completely neglect this structural information. Nonetheless, it has long been known that certain isoforms are selective based on general small molecule characteristics, such as planarity, charge, etc. These empirical preferences suggest that understanding the shape, flexibility, and electrostatic nature of the CYP binding pockets can lead to improved modeling of CYP metabolism vis-à-vis approaches that only consider properties of the small molecules themselves.

In this work, both 2D methods and 3D methods are used to predict the isoform selectivity, small molecule reactivity, and regioselectivity of CYPs. The 2D-based methods developed herein are parsimonious yet accurate and can be used to quickly evaluate selectivity and reactivity. The 3D approach is based on pharmacophore modeling, which provides a rapid and flexible way to predict CYP isoform selectivity and regioselectivity. The modular components of the pharmacophore afford a straightforward means to tailor for a particular problem of interest, thereby improving prediction accuracy. Moreover, directly incorporating 3D CYP structures into the models confers unique advantages over 2D-based approaches, such as the ability to distinguish reactivity differences among stereoisomers. Finally, predicted results can readily be visualized, and thus potential CYP liabilities are not merely evaluated or flagged, but can also be designed against – a clear departure from the pass/fail filtering paradigm prevalent in most CYP modeling efforts to date.

While the orally active azoles fluconazole, posaconazole, and isavuconazole are effective antifungal agents, they exhibit a myriad of off-target issues such as drug-drug interactions, anaphylaxis, and liver and reproductive toxicities, mainly due to their inhibition of other P-450 enzymes. We have described the discovery and development of the more selective tetrazole antifungal agents VT-1161 and VT-1598, which exhibit this increase in specificity while maintaining high potency. In an effort to find equally selective and potent compounds with different physical properties, we investigated analogues with alternative backbone scaffolds which would give us an improvement in PK/PD profiles as well as provide possible functionality for creation of prodrugs.

The pregnane X receptor (PXR) is a promiscuous nuclear hormone receptor responsible for many undesirable drug interactions. This study aims to describe the minimum ligand components that are required for PXR interaction and the features that differentiate actives from inactives, especially with regard to ligand geometric shapes, electrostatic distributions, and distances. A novel molecular descriptor, Smallest Maximum Interatomic Distance (SMID), is described, together with strategies for designing analogs with increased SMID. An interaction model is developed and evaluated against a temporal test set.

We have identified a novel PDE2 inhibitor series using fragment-based screening. Pyrazolopyrimidine, while possessing weak potency (K_i = 22.4 μM), exhibited good binding efficiencies (LBE = 0.49, LLE = 4.48) to serve as a start for structure-based drug design. Using structure-based approach, this fragment was developed into a series of potent PDE2 inhibitors with good physicochemical properties. A PDE2 selective inhibitor was identified that exhibited favorable rat pharmacokinetic properties.