RECOVER identifies synergistic drug combinations in vitro through sequential model optimization.

For large libraries of small molecules, exhaustive combinatorial chemical screens become infeasible to perform when considering a range of disease models, assay conditions, and dose ranges. Deep learning models have achieved state-of-the-art results in silico for the prediction of synergy scores. However, databases of drug combinations are biased toward synergistic agents and results do not generalize out of distribution. During 5 rounds of experimentation, we employ sequential model optimization with a deep learning model to select drug combinations increasingly enriched for synergism and active against a cancer cell line-evaluating only ∼5% of the total search space. Moreover, we find that learned drug embeddings (using structural information) begin to reflect biological mechanisms. In silico benchmarking suggests search queries are ∼5-10× enriched for highly synergistic drug combinations by using sequential rounds of evaluation when compared with random selection or ∼3× when using a pretrained model.

Validation of a small molecule inhibitor of PDE6D-RAS interaction with favorable anti-leukemic effects.

RAS mutations prevalent in high-risk leukemia have been linked to relapse and chemotherapy resistance. Efforts to directly target RAS proteins have been largely unsuccessful. However, since RAS-mediated transformation is dependent on signaling through the RAS-related C3 botulinum toxin substrate (RAC) small GTPase, we hypothesized that targeting RAC may be an effective therapeutic approach in RAS mutated tumors. Here we describe multiple small molecules capable of inhibiting RAC activation in acute lymphoblastic leukemia cell lines. One of these, DW0254, also demonstrates promising anti-leukemic activity in RAS-mutated cells. Using chemical proteomics and biophysical methods, we identified the hydrophobic pocket of phosphodiester 6 subunit delta (PDE6D), a known RAS chaperone, as a target for this compound. Inhibition of RAS localization to the plasma membrane upon DW0254 treatment is associated with RAC inhibition through a phosphatidylinositol-3-kinase/AKT-dependent mechanism. Our findings provide new insights into the importance of PDE6D-mediated transport for RAS-dependent RAC activation and leukemic cell survival.

A human-based multi-gene signature enables quantitative drug repurposing for metabolic disease.

Insulin resistance (IR) contributes to the pathophysiology of diabetes, dementia, viral infection, and cardiovascular disease. Drug repurposing (DR) may identify treatments for IR; however, barriers include uncertainty whether in vitro transcriptomic assays yield quantitative pharmacological data, or how to optimise assay design to best reflect in vivo human disease. We developed a clinical-based human tissue IR signature by combining lifestyle-mediated treatment responses (>500 human adipose and muscle biopsies) with biomarkers of disease status (fasting IR from >1200 biopsies). The assay identified a chemically diverse set of >130 positively acting compounds, highly enriched in true positives, that targeted 73 proteins regulating IR pathways. Our multi-gene RNA assay score reflected the quantitative pharmacological properties of a set of epidermal growth factor receptor-related tyrosine kinase inhibitors, providing insight into drug target specificity; an observation supported by deep learning-based genome-wide predicted pharmacology. Several drugs identified are suitable for evaluation in patients, particularly those with either acute or severe chronic IR.

Developing a new drug that is both safe and effective is a complex and expensive endeavor. An alternative approach is to 'repurpose' existing, safe compounds ­ that is, to establish if they could treat conditions others than the ones they were initially designed for. To achieve this, methods that can predict the activity of thousands of established drugs are necessary. These approaches are particularly important for conditions for which it is hard to find promising treatment. This includes, for instance, heart failure, dementia and other diseases that are linked to the activity of the hormone insulin becoming modified throughout the body, a defect called insulin resistance. Unfortunately, it is difficult to model the complex actions of insulin using cells in the lab, because they involve intricate networks of proteins, tissues and metabolites. Timmons et al. set out to develop a way to better assess whether a drug could be repurposed to treat insulin resistance. The aim was to build a biological signature of the disease in multiple human tissues, as this would help to make the findings more relevant to the clinic. This involved examining which genes were switched on or off in thousands of tissue samples from patients with different degrees of insulin resistance. Importantly, some of the patients had their condition reversed through lifestyle changes, while others did not respond well to treatment. These 'non-responders' provided crucial new clues to screen for active drugs. Carefully piecing the data together revealed the molecules and pathways most related to the severity of insulin resistance. Cross-referencing these results with the way existing drugs act on gene activity, highlighted 138 compounds that directly bind 73 proteins responsible for regulating insulin resistance pathways. Some of the drugs identified are suitable for short-term clinical studies, and it may even be possible to rank similar compounds based on their chemical activity. Beyond giving a glimpse into the complex molecular mechanisms of insulin resistance in humans, Timmons et al. provide a fresh approach to how drugs could be repurposed, which could be adapted to other conditions.

Deep Learning in Structure-Based Drug Design.

Computational methods play an increasingly important role in drug discovery. Structure-based drug design (SBDD), in particular, includes techniques that take into account the structure of the macromolecular target to predict compounds that are likely to establish optimal interactions with the binding site. The current interest in machine learning algorithms based on deep neural networks encouraged the application of deep learning to SBDD related problems. This chapter covers selected works in this active area of research.

Learning protein-ligand binding affinity with atomic environment vectors.

Scoring functions for the prediction of protein-ligand binding affinity have seen renewed interest in recent years when novel machine learning and deep learning methods started to consistently outperform classical scoring functions. Here we explore the use of atomic environment vectors (AEVs) and feed-forward neural networks, the building blocks of several neural network potentials, for the prediction of protein-ligand binding affinity. The AEV-based scoring function, which we term AEScore, is shown to perform as well or better than other state-of-the-art scoring functions on binding affinity prediction, with an RMSE of 1.22 pK units and a Pearson's correlation coefficient of 0.83 for the CASF-2016 benchmark. However, AEScore does not perform as well in docking and virtual screening tasks, for which it has not been explicitly trained. Therefore, we show that the model can be combined with the classical scoring function AutoDock Vina in the context of [Formula: see text]-learning, where corrections to the AutoDock Vina scoring function are learned instead of the protein-ligand binding affinity itself. Combined with AutoDock Vina, [Formula: see text]-AEScore has an RMSE of 1.32 pK units and a Pearson's correlation coefficient of 0.80 on the CASF-2016 benchmark, while retaining the docking and screening power of the underlying classical scoring function.

Enhancing Chemogenomics with Predictive Pharmacology.

One of the grand challenges in contemporary chemical biology is the generation of a probe for every member of the human proteome. Probe selection and optimization strategies typically rely on experimental bioactivity data to determine the potency and selectivity of candidate molecules. However, this approach is profoundly limited by the sparsity of the known data, the annotation bias often found in the literature, and the cost of physical screening. Recent advancements in predictive pharmacology, such as the application of multitask and transfer learning, as well as the use of biologically motivated, structure-agnostic features to characterize molecules, should serve to mitigate these issues. Computational modeling likely offers the only cost-effective approach to substantially increasing the bioactivity annotation density both on the local and global scale and thus, we argue, will need to make a substantial contribution if the ambitious goals of probing the human proteome are to be realized in the foreseeable future.

Underappreciated Chemical Interactions in Protein-Ligand Complexes.

Non-covalent interactions lie at the bases of the molecular recognition process. In medicinal chemistry, understanding how bioactive molecules interact with their target can help to explain structure-activity relationships (SAR) and improve potency of lead compounds. In particular, computational analysis of protein-ligand complexes can help to unravel key interactions and guide structure-based drug design.The literature describing protein-ligand complexes is typically focused on few types of non-covalent interactions (e.g., hydrophobic contacts, hydrogen bonds, and salt bridges). Stacking interactions involving aromatic rings are also relatively well known to medicinal chemistry practitioners. Potency optimization efforts are often focused on targeting these interactions. However, a variety of underappreciated interactions were shown to have a relevant effect on the stabilization of protein-ligand complexes. This chapter aims at listing selected non-covalent interactions and discuss some examples on how they can impact drug design.

A Hybrid Virtual Screening Protocol Based on Binding Mode Similarity.

In structure-based virtual screening (SBVS), a scoring function is usually applied to rank a database of docked compounds. Docking programs are often successful in reproducing experimental binding modes; however, the estimation of binding affinity still is the Achilles' heel of docking. The integration of SB and ligand-based (LB) methods is considered a promising strategy to increase hit rates in VS. Herein, we describe a hybrid protocol that is based on the assessment of binding mode similarity between docked compounds and a bound reference ligand. In this context, both experimental and computationally modeled poses have been successfully used as references for three-dimensional (3D) similarity calculations. In this chapter, the methods applied in recent validation studies are described.

Identification of 4-aryl-1H-pyrrole[2,3-b]pyridine derivatives for the development of new B-Raf inhibitors.

During the last years, a significant interest in the identification of new classes of B-Raf inhibitors has emerged. In this study, which was conceived within an effort that culminated in the recent report of the first dual inhibitors of B-Raf and Hsp90, we describe the identification of four compounds based on 4-aryl-1H-pyrrole[2,3-b]pyridine scaffold as interesting starting points for the development of new B-Raf inhibitors. Structure-activity relationships and predicted binding modes are discussed. Moreover, the novelty of the newly identified structures with respect to currently known B-Raf inhibitors was assessed through a ligand-based dissimilarity assessment. Finally, structural modifications with the potential ability to improve the activity toward B-Raf are put forward.

Compound Ranking Based on Fuzzy Three-Dimensional Similarity Improves the Performance of Docking into Homology Models of G-Protein-Coupled Receptors.

Ligand docking into homology models of G-protein-coupled receptors (GPCRs) is a widely used approach in computational compound screening. The generation of "double-hypothetical" models of ligand-target complexes has intrinsic accuracy limitations that further complicate compound ranking and selection compared to those of X-ray structures. Given these uncertainties, we have explored "fuzzy 3D similarity" between hypothetical binding modes of known ligands in homology models and docking poses of database compounds as an alternative to conventional scoring schemes. Therefore, GPCR homology models at varying accuracy levels were generated and used for docking. Increases in recall performance were observed for fuzzy 3D similarity ranking using single or multiple ligand poses compared to that of conventional scoring functions and interaction fingerprints. Fuzzy similarity ranking was also successfully applied to docking into an external model of a GPCR for which no experimental structure is currently available. Taken together, our results indicate that the use of putative ligand poses, albeit approximate at best, increases the odds of identifying active compounds in docking screens of GPCR homology models.

Predicting bioactive conformations and binding modes of macrocycles.

Macrocyclic compounds experience increasing interest in drug discovery. It is often thought that these large and chemically complex molecules provide promising candidates to address difficult targets and interfere with protein-protein interactions. From a computational viewpoint, these molecules are difficult to treat. For example, flexible docking of macrocyclic compounds is hindered by the limited ability of current docking approaches to optimize conformations of extended ring systems for pose prediction. Herein, we report predictions of bioactive conformations of macrocycles using conformational search and binding modes using docking. Conformational ensembles generated using specialized search technique of about 70 % of the tested macrocycles contained accurate bioactive conformations. However, these conformations were difficult to identify on the basis of conformational energies. Moreover, docking calculations with limited ligand flexibility starting from individual low energy conformations rarely yielded highly accurate binding modes. In about 40 % of the test cases, binding modes were approximated with reasonable accuracy. However, when conformational ensembles were subjected to rigid body docking, an increase in meaningful binding mode predictions to more than 50 % of the test cases was observed. Electrostatic effects did not contribute to these predictions in a positive or negative manner. Rather, achieving shape complementarity at macrocycle-target interfaces was a decisive factor. In summary, a combined computational protocol using pre-computed conformational ensembles of macrocycles as a starting point for docking shows promise in modeling binding modes of macrocyclic compounds.

Binding mode similarity measures for ranking of docking poses: a case study on the adenosine A2A receptor.

We report an investigation designed to explore alternative approaches for ranking of docking poses in the search for antagonists of the adenosine A2A receptor, an attractive target for structure-based virtual screening. Calculation of 3D similarity of docking poses to crystallographic ligand(s) as well as similarity of receptor-ligand interaction patterns was consistently superior to conventional scoring functions for prioritizing antagonists over decoys. Moreover, the use of crystallographic antagonists and agonists, a core fragment of an antagonist, and a model of an agonist placed into the binding site of an antagonist-bound form of the receptor resulted in a significant early enrichment of antagonists in compound rankings. Taken together, these findings showed that the use of binding modes of agonists and/or antagonists, even if they were only approximate, for similarity assessment of docking poses or comparison of interaction patterns increased the odds of identifying new active compounds over conventional scoring.

Three-Dimensional Similarity in Molecular Docking: Prioritizing Ligand Poses on the Basis of Experimental Binding Modes.

Molecular docking is the premier approach to structure-based virtual screening. While ligand posing is often successful, compound ranking using force field-based scoring functions remains difficult. Uncertainties associated with scoring often limit the ability to confidently identify new active compounds. In this study, we introduce an alternative approach to compound ranking. Rather than using scoring functions for final ranking, compounds are prioritized on the basis of computed 3D similarity to known crystallographic ligands. For different targets, it is shown that 3D similarity-based ranking consistently improves the enrichment of active compounds compared to ranking obtained using scoring functions, even if only a single crystallographic ligand is used as a reference. While the strategy is not applicable in cases where no cocrystal structure is available, it should be a promising alternative or complement to conventional scoring in many instances. Since ligand similarity calculations are used to rank docking poses, which are independently derived, the approach introduced herein also contributes to the integration of ligand- and structure-based computational screening methods.

Insights into the interaction of negative allosteric modulators with the metabotropic glutamate receptor 5: discovery and computational modeling of a new series of ligands with nanomolar affinity.

Metabotropic glutamate receptor 5 (mGlu5) is a biological target implicated in major neurological and psychiatric disorders. In the present study, we have investigated structural determinants of the interaction of negative allosteric modulators (NAMs) with the seven-transmembrane (7TM) domain of mGlu5. A homology model of the 7TM receptor domain built on the crystal structure of the mGlu1 template was obtained, and the binding modes of known NAMs, namely MPEP and fenobam, were investigated by docking and molecular dynamics simulations. The results were validated by comparison with mutagenesis data available in the literature for these two ligands, and subsequently corroborated by the recently described mGlu5 crystal structure. Moreover, a new series of NAMs was synthesized and tested, providing compounds with nanomolar affinity. Several structural modifications were sequentially introduced with the aim of identifying structural features important for receptor binding. The synthesized NAMs were docked in the validated homology model and binding modes were used to interpret and discuss structure-activity relationships within this new series of compounds. Finally, the models of the interaction of NAMs with mGlu5 were extended to include important non-aryl alkyne mGlu5 NAMs taken from the literature. Overall, the results provide useful insights into the molecular interaction of negative allosteric modulators with mGlu5 and may facilitate the design of new modulators for this class of receptors.

Computational polypharmacology analysis of the heat shock protein 90 interactome.

The design of a single drug molecule that is able to simultaneously and specifically interact with multiple biological targets is gaining major consideration in drug discovery. However, the rational design of drugs with a desired polypharmacology profile is still a challenging task, especially when these targets are distantly related or unrelated. In this work, we present a computational approach aimed at the identification of suitable target combinations for multitarget drug design within an ensemble of biologically relevant proteins. The target selection relies on the analysis of activity annotations present in molecular databases and on ligand-based virtual screening. A few target combinations were also inspected with structure-based methods to demonstrate that the identified dual-activity compounds are able to bind target combinations characterized by remote binding site similarities. Our approach was applied to the heat shock protein 90 (Hsp90) interactome, which contains several targets of key importance in cancer. Promising target combinations were identified, providing a basis for the computational design of compounds with dual activity. The approach may be used on any ensemble of proteins of interest for which known inhibitors are available.

Role of bifidobacteria in the hydrolysis of chlorogenic acid.

This study aimed to explore the capability of potentially probiotic bifidobacteria to hydrolyze chlorogenic acid into caffeic acid (CA), and to recognize the enzymes involved in this reaction. Bifidobacterium strains belonging to eight species occurring in the human gut were screened. The hydrolysis seemed peculiar of Bifidobacterium animalis, whereas the other species failed to release CA. Intracellular feruloyl esterase activity capable of hydrolyzing chlorogenic acid was detected only in B. animalis. In silico research among bifidobacteria esterases identified Balat_0669 as the cytosolic enzyme likely responsible of CA release in B. animalis. Comparative modeling of Balat_0669 and molecular docking studies support its role in chlorogenic acid hydrolysis. Expression, purification, and functional characterization of Balat_0669 in Escherichia coli were obtained as further validation. A possible role of B. animalis in the activation of hydroxycinnamic acids was demonstrated and new perspectives were opened in the development of new probiotics, specifically selected for the enhanced bioconversion of phytochemicals into bioactive compounds.

Structure-based discovery of the first allosteric inhibitors of cyclin-dependent kinase 2.

Allosteric targeting of protein kinases via displacement of the structural αC helix with type III allosteric inhibitors is currently gaining a foothold in drug discovery. Recently, the first crystal structure of CDK2 with an open allosteric pocket adjacent to the αC helix has been described, prospecting new opportunities to design more selective inhibitors, but the structure has not yet been exploited for the structure-based design of type III allosteric inhibitors. In this work we report the results of a virtual screening campaign that resulted in the discovery of the first-in-class type III allosteric ligands of CDK2. Using a combination of docking and post-docking analyses made with our tool BEAR, 7 allosteric ligands (hit rate of 20%) with micromolar affinity for CDK2 were identified, some of them inhibiting the growth of breast cancer cell lines in the micromolar range. Competition experiments performed in the presence of the ATP-competitive inhibitor staurosporine confirmed that the 7 ligands are truly allosteric, in agreement with their design. Of these, compound 2 bound CDK2 with an EC50 value of 3 µM and inhibited the proliferation of MDA-MB231 and ZR-75-1 breast cancer cells with IC50 values of approximately 20 µM, while compound 4 had an EC50 value of 71 µM and IC50 values around 4 µM. Remarkably, the most potent compound 4 was able to selectively inhibit CDK2-mediated Retinoblastoma phosphorylation, confirming that its mechanism of action is fully compatible with a selective inhibition of CDK2 phosphorylation in cells. Finally, hit expansion through analog search of the most potent inhibitor 4 revealed an additional ligand 4g with similar in vitro potency on breast cancer cells.

Polypharmacology: challenges and opportunities in drug discovery.

At present, the legendary magic bullet, i.e., a drug with high potency and selectivity toward a specific biological target, shares the spotlight with an emerging and alternative polypharmacology approach. Polypharmacology suggests that more effective drugs can be developed by specifically modulating multiple targets. It is generally thought that complex diseases such as cancer and central nervous system diseases may require complex therapeutic approaches. In this respect, a drug that "hits" multiple sensitive nodes belonging to a network of interacting targets offers the potential for higher efficacy and may limit drawbacks generally arising from the use of a single-target drug or a combination of multiple drugs. In this review, we will compare advantages and disadvantages of multitarget versus combination therapies, discuss potential drug promiscuity arising from off-target effects, comment on drug repurposing, and introduce approaches to the computational design of multitarget drugs.

Enrichment factor analyses on G-protein coupled receptors with known crystal structure.

G-protein coupled receptors (GPCRs) are highly relevant drug targets. Four GPCRs with known crystal structure were analyzed with docking (AutoDock4) and postdocking (MM-PBSA) in order to evaluate the ability to recognize known antagonists from a larger database of molecular decoys and to predict correct binding modes. Moreover, implications on multitarget drug screening are put forward. The results suggest that these methods may be of interest to the growing field of GPCR structure-based virtual screening.

Application of a post-docking procedure based on MM-PBSA and MM-GBSA on single and multiple protein conformations.

In the last decades, molecular docking has emerged as an increasingly useful tool in the modern drug discovery process, but it still needs to overcome many hurdles and limitations such as how to account for protein flexibility and poor scoring function performance. For this reason, it has been recognized that in many cases docking results need to be post-processed to achieve a significant agreement with experimental activities. In this study, we have evaluated the performance of MM-PBSA and MM-GBSA scoring functions, implemented in our post-docking procedure BEAR, in rescoring docking solutions. For the first time, the performance of this post-docking procedure has been evaluated on six different biological targets (namely estrogen receptor, thymidine kinase, factor Xa, adenosine deaminase, aldose reductase, and enoyl ACP reductase) by using i) both a single and a multiple protein conformation approach, and ii) two different software, namely AutoDock and LibDock. The assessment has been based on two of the most important criteria for the evaluation of docking methods, i.e., the ability of known ligands to enrich the top positions of a ranked database with respect to molecular decoys, and the consistency of the docking poses with crystallographic binding modes. We found that, in many cases, MM-PBSA and MM-GBSA are able to yield higher enrichment factors compared to those obtained with the docking scoring functions alone. However, for only a minority of the cases, the enrichment factors obtained by using multiple protein conformations were higher than those obtained by using only one protein conformation.