Application of AlphaFold models in evaluating ligandable cysteines across E3 ligases.

Proteolysis Targeting Chimeras (PROTACs) are an emerging therapeutic modality and chemical biology tools for Targeted Protein Degradation (TPD). PROTACs contain a ligand targeting the protein of interest, a ligand recruiting an E3 ligase and a linker connecting these two ligands. There are over 600 E3 ligases known so far, but only a handful have been exploited for TPD applications. A key reason for this is the scarcity of ligands binding various E3 ligases and the paucity of structural data available, which complicates ligand design across the family. In this study, we aim to progress PROTAC discovery by proposing a shortlist of E3 ligases that can be prioritized for covalent targeting by performing systematic structural ligandability analysis on a chemoproteomic dataset of potentially reactive cysteines across hundreds of E3 ligases. One of the goals of this study is to apply AlphaFold (AF) models for ligandability evaluations, as for a vast majority of these ligases an experimental structure is not available in the protein data bank (PDB). Using a combination of pocket features, AF model quality and additional aspects, we propose a shortlist of E3 ligases and corresponding cysteines that can be prioritized to potentially discover covalent ligands and expand the PROTAC toolbox.

Divide and Conquer. Pocket-Opening Mixed-Solvent Simulations in the Perspective of Docking Virtual Screening Applications for Drug Discovery.

The existence of a druggable binding pocket is a prerequisite for computational drug-target interaction studies including virtual screening. Retrospective studies have shown that extended sampling methods like Markov State Modeling and mixed-solvent simulations can identify cryptic pockets relevant for drug discovery. Here, we apply a combination of mixed-solvent molecular dynamics (MD) and time-structure independent component analysis (TICA) to four retrospective case studies: NPC2, the CECR2 bromodomain, TEM-1, and MCL-1. We compare previous experimental and computational findings to our results. It is shown that the successful identification of cryptic pockets depends on the system and the cosolvent probes. We used alternative TICA internal features such as the unbiased backbone coordinates or backbone dihedrals versus biased interatomic distances. We found that in the case of NPC2, TEM-1, and MCL-1, the use of unbiased features is able to identify cryptic pockets, although in the case of the CECR2 bromodomain, more specific features are required to properly capture a pocket opening. In the perspective of virtual screening applications, it is shown how docking studies with the parent ligands depend critically on the conformational state of the targets.

Annotation of Allosteric Compounds to Enhance Bioactivity Modeling for Class A GPCRs.

Proteins often have both orthosteric and allosteric binding sites. Endogenous ligands, such as hormones and neurotransmitters, bind to the orthosteric site, while synthetic ligands may bind to orthosteric or allosteric sites, which has become a focal point in drug discovery. Usually, such allosteric modulators bind to a protein noncompetitively with its endogenous ligand or substrate. The growing interest in allosteric modulators has resulted in a substantial increase of these entities and their features such as binding data in chemical libraries and databases. Although this data surge fuels research focused on allosteric modulators, binding data is unfortunately not always clearly indicated as being allosteric or orthosteric. Therefore, allosteric binding data is difficult to retrieve from databases that contain a mixture of allosteric and orthosteric compounds. This decreases model performance when statistical methods, such as machine learning models, are applied. In previous work we generated an allosteric data subset of ChEMBL release 14. In the current study an improved text mining approach is used to retrieve the allosteric and orthosteric binding types from the literature in ChEMBL release 22. Moreover, convolutional deep neural networks were constructed to predict the binding types of compounds for class A G protein-coupled receptors (GPCRs). Temporal split validation showed the model predictiveness with Matthews correlation coefficient (MCC) = 0.54, sensitivity allosteric = 0.54, and sensitivity orthosteric = 0.94. Finally, this study shows that the inclusion of accurate binding types increases binding predictions by including them as descriptor (MCC = 0.27 improved to MCC = 0.34; validated for class A GPCRs, trained on all GPCRs). Although the focus of this study is mainly on class A GPCRs, binding types for all protein classes in ChEMBL were obtained and explored. The data set is included as a supplement to this study, allowing the reader to select the compounds and binding types of interest.

Accurate Prediction of GPCR Ligand Binding Affinity with Free Energy Perturbation.

The computational prediction of relative binding free energies is a crucial goal for drug discovery, and G protein-coupled receptors (GPCRs) are arguably the most important drug target class. However, they present increased complexity to model compared to soluble globular proteins. Despite breakthroughs, experimental X-ray crystal and cryo-EM structures are challenging to attain, meaning computational models of the receptor and ligand binding mode are sometimes necessary. This leads to uncertainty in understanding ligand-protein binding induced changes such as, water positioning and displacement, side chain positioning, hydrogen bond networks, and the overall structure of the hydration shell around the ligand and protein. In other words, the very elements that define structure activity relationships (SARs) and are crucial for accurate binding free energy calculations are typically more uncertain for GPCRs. In this work we use free energy perturbation (FEP) to predict the relative binding free energies for ligands of two different GPCRs. We pinpoint the key aspects for success such as the important role of key water molecules, amino acid ionization states, and the benefit of equilibration with specific ligands. Initial calculations following typical FEP setup and execution protocols delivered no correlation with experiment, but we show how results are improved in a logical and systematic way. This approach gave, in the best cases, a coefficient of determination (R2) compared with experiment in the range of 0.6-0.9 and mean unsigned errors compared to experiment of 0.6-0.7 kcal/mol. We anticipate that our findings will be applicable to other difficult-to-model protein ligand data sets and be of wide interest for the community to continue improving FE binding energy predictions.

Successive Statistical and Structure-Based Modeling to Identify Chemically Novel Kinase Inhibitors.

Kinases are frequently studied in the context of anticancer drugs. Their involvement in cell responses, such as proliferation, differentiation, and apoptosis, makes them interesting subjects in multitarget drug design. In this study, a workflow is presented that models the bioactivity spectra for two panels of kinases: (1) inhibition of RET, BRAF, SRC, and S6K, while avoiding inhibition of MKNK1, TTK, ERK8, PDK1, and PAK3, and (2) inhibition of AURKA, PAK1, FGFR1, and LKB1, while avoiding inhibition of PAK3, TAK1, and PIK3CA. Both statistical and structure-based models were included, which were thoroughly benchmarked and optimized. A virtual screening was performed to test the workflow for one of the main targets, RET kinase. This resulted in 5 novel and chemically dissimilar RET inhibitors with remaining RET activity of <60% (at a concentration of 10 µM) and similarities with known RET inhibitors from 0.18 to 0.29 (Tanimoto, ECFP6). The four more potent inhibitors were assessed in a concentration range and proved to be modestly active with a pIC50 value of 5.1 for the most active compound. The experimental validation of inhibitors for RET strongly indicates that the multitarget workflow is able to detect novel inhibitors for kinases, and hence, this workflow can potentially be applied in polypharmacology modeling. We conclude that this approach can identify new chemical matter for existing targets. Moreover, this workflow can easily be applied to other targets as well.

Advances and Challenges in Computational Target Prediction.

Target deconvolution is a vital initial step in preclinical drug development to determine research focus and strategy. In this respect, computational target prediction is used to identify the most probable targets of an orphan ligand or the most similar targets to a protein under investigation. Applications range from the fundamental analysis of the mode-of-action over polypharmacology or adverse effect predictions to drug repositioning. Here, we provide a review on published ligand- and target-based as well as hybrid approaches for computational target prediction, together with current limitations and future directions.

Chemical space screening around Phe3 in opioid peptides: Modulating µ versus δ agonism by Suzuki-Miyaura cross-couplings.

In this study, affinities and activities of derivatized analogues of Dmt-dermorphin[1-4] (i.e. Dmt-d-Ala-Phe-GlyNH2, Dmt = 2',6'-dimethyl-(S)-tyrosine) for the µ opioid receptor (MOP) and δ opioid receptor (DOP) were evaluated using radioligand binding studies, functional cell-based assays and isolated organ bath experiments. By means of solid-phase or solution-phase Suzuki-Miyaura cross-couplings, various substituted regioisomers of the phenylalanine moiety in position 3 of the sequence were prepared. An 18-membered library of opioid tetrapeptides was generated via screening of the chemical space around the Phe3 side chain. These substitutions modulated bioactivity, receptor subtype selectivity and highly effective ligands with subnanomolar binding affinities, contributed to higher functional activities and potent analgesic actions. In search of selective peptidic ligands, we show here that the Suzuki-Miyaura reaction is a versatile and robust tool which could also be deployed elsewhere.

Identification of Allosteric Modulators of Metabotropic Glutamate 7 Receptor Using Proteochemometric Modeling.

Proteochemometric modeling (PCM) is a computational approach that can be considered an extension of quantitative structure-activity relationship (QSAR) modeling, where a single model incorporates information for a family of targets and all the associated ligands instead of modeling activity versus one target. This is especially useful for situations where bioactivity data exists for similar proteins but is scarce for the protein of interest. Here we demonstrate the application of PCM to identify allosteric modulators of metabotropic glutamate (mGlu) receptors. Given our long-running interest in modulating mGlu receptor function we compiled a matrix of compound-target bioactivity data. Some members of the mGlu family are well explored both internally and in the public domain, while there are much fewer examples of ligands for other targets such as the mGlu7 receptor. Using a PCM approach mGlu7 receptor hits were found. In comparison to conventional single target modeling the identified hits were more diverse, had a better confirmation rate, and provide starting points for further exploration. We conclude that the robust structure-activity relationship from well explored target family members translated to better quality hits for PCM compared to virtual screening (VS) based on a single target.

Interacting with GPCRs: Using Interaction Fingerprints for Virtual Screening.

The expanding number of crystal structures of G protein-coupled receptors (GPCRs) has increased the knowledge on receptor function and their ability to recognize ligands. Although structure-based virtual screening has been quite successful on GPCRs, scores obtained by docking are typically not indicative for ligand affinity. Methods capturing interactions between protein and ligand in a more explicit manner, such as interaction fingerprints (IFPs), have been applied as an addition or alternative to docking. Originally IFPs captured the interactions of amino acid residues with ligands with specific definitions for the various interaction types. More complex IFPs now capture atom-atom interactions, such as in SYBYL, or fragment-fragment co-occurrences such as in SPLIF. Overall, most of the IFPs have been studied in comparison with docking in retrospective studies. For GPCRs it remains unclear which IFP should be used, if at all, and in what manner. Thus, the performance between five different IFPs was compared on five different representative GPCRs, including several extensions of the original implementations,. Results show that the more detailed IFPs, SYBYL and SPLIF, perform better than the other IFPs (Deng, Credo, and Elements). SPLIF was further tuned based on the number of poses, fingerprint similarity coefficient, and using an ensemble of structures. Enrichments were obtained that were significantly higher than initial enrichments and those obtained by 2D-similarity. With the increase in available crystal structures for GPCRs, and given that IFPs such as SPLIF enhance enrichment in virtual screens, it is anticipated that IFPs will be used in conjunction with docking, especially for GPCRs with a large binding pocket.

In search of novel ligands using a structure-based approach: a case study on the adenosine A2A receptor.

In this work, we present a case study to explore the challenges associated with finding novel molecules for a receptor that has been studied in depth and has a wealth of chemical information available. Specifically, we apply a previously described protocol that incorporates explicit water molecules in the ligand binding site to prospectively screen over 2.5 million drug-like and lead-like compounds from the commercially available eMolecules database in search of novel binders to the adenosine A2A receptor (A2AAR). A total of seventy-one compounds were selected for purchase and biochemical assaying based on high ligand efficiency and high novelty (Tanimoto coefficient ≤0.25 to any A2AAR tested compound). These molecules were then tested for their affinity to the adenosine A2A receptor in a radioligand binding assay. We identified two hits that fulfilled the criterion of ~50 % radioligand displacement at a concentration of 10 µM. Next we selected an additional eight novel molecules that were predicted to make a bidentate interaction with Asn2536.55, a key interacting residue in the binding pocket of the A2AAR. None of these eight molecules were found to be active. Based on these results we discuss the advantages of structure-based methods and the challenges associated with finding chemically novel molecules for well-explored targets.

Antibody humanization by redesign of complementarity-determining region residues proximate to the acceptor framework.

Antibodies are key components of the adaptive immune system and are well-established protein therapeutic agents. Typically high-affinity antibodies are obtained by immunization of rodent species that need to be humanized to reduce their immunogenicity. The complementarity-determining regions (CDRs) contain the residues in a defined loop structure that confer antigen binding, which must be retained in the humanized antibody. To design a humanized antibody, we graft the mature murine CDRs onto a germline human acceptor framework. Structural defects due to mismatches at the graft interface can be fixed by mutating some framework residues to murine, or by mutating some residues on the CDRs' backside to human or to a de novo designed sequence. The first approach, framework redesign, can yield an antibody with binding better than the CDR graft and one equivalent to the mature murine, and reduced immunogenicity. The second approach, CDR redesign, is presented here as a new approach, yielding an antibody with binding better than the CDR graft, and immunogenicity potentially less than that from framework redesign. Application of both approaches to the humanization of anti-α4 integrin antibody HP1/2 is presented and the concept of the hybrid humanization approach that retains "difficult to match" murine framework amino acids and uses de novo CDR design to minimize murine amino acid content and reduce cell-mediated cytotoxicity liabilities is discussed.

Influence of canonical structure determining residues on antibody affinity and stability.

A number of light and heavy chain canonical residue core redesigns were made in a therapeutic antibody (AQC2, anti-VLA1) Fab to explore the consequences to binding affinity and stability. These positions are all loop supporting, primarily CDR1 residues which do not directly contact the antigen. Structure based methods were used with and without consensus sequence information. 30 constructs were made, 24 expressed, and 70% of the designs using consensus sequence information retained binding affinity. Some success maintaining stability with more extreme redesigns suggests a surprising tolerance to mutation, though it often comes at the cost of loss of binding affinity and presumed loop conformation changes. In concordance with the expected need to present an ordered surface for binding, a relationship between decreased affinity and decreased stability was observed. Overpacking the core tends to destabilize the molecule and should be avoided.

Significantly improved HIV inhibitor efficacy prediction employing proteochemometric models generated from antivirogram data.

Infection with HIV cannot currently be cured; however it can be controlled by combination treatment with multiple anti-retroviral drugs. Given different viral genotypes for virtually each individual patient, the question now arises which drug combination to use to achieve effective treatment. With the availability of viral genotypic data and clinical phenotypic data, it has become possible to create computational models able to predict an optimal treatment regimen for an individual patient. Current models are based only on sequence data derived from viral genotyping; chemical similarity of drugs is not considered. To explore the added value of chemical similarity inclusion we applied proteochemometric models, combining chemical and protein target properties in a single bioactivity model. Our dataset was a large scale clinical database of genotypic and phenotypic information (in total ca. 300,000 drug-mutant bioactivity data points, 4 (NNRTI), 8 (NRTI) or 9 (PI) drugs, and 10,700 (NNRTI) 10,500 (NRTI) or 27,000 (PI) mutants). Our models achieved a prediction error below 0.5 Log Fold Change. Moreover, when directly compared with previously published sequence data, derived models PCM performed better in resistance classification and prediction of Log Fold Change (0.76 log units versus 0.91). Furthermore, we were able to successfully confirm both known and identify previously unpublished, resistance-conferring mutations of HIV Reverse Transcriptase (e.g. K102Y, T216M) and HIV Protease (e.g. Q18N, N88G) from our dataset. Finally, we applied our models prospectively to the public HIV resistance database from Stanford University obtaining a correct resistance prediction rate of 84% on the full set (compared to 80% in previous work on a high quality subset). We conclude that proteochemometric models are able to accurately predict the phenotypic resistance based on genotypic data even for novel mutants and mixtures. Furthermore, we add an applicability domain to the prediction, informing the user about the reliability of predictions.

Selecting an optimal number of binding site waters to improve virtual screening enrichments against the adenosine A2A receptor.

A major challenge in structure-based virtual screening (VS) involves the treatment of explicit water molecules during docking in order to improve the enrichment of active compounds over decoys. Here we have investigated this in the context of the adenosine A2A receptor, where water molecules have previously been shown to be important for achieving high enrichment rates with docking, and where the positions of some binding site waters are known from a high-resolution crystal structure. The effect of these waters (both their presence and orientations) on VS enrichment was assessed using a carefully curated set of 299 high affinity A2A antagonists and 17,337 decoys. We show that including certain crystal waters greatly improves VS enrichment and that optimization of water hydrogen positions is needed in order to achieve the best results. We also show that waters derived from a molecular dynamics simulation - without any knowledge of crystallographic waters - can improve enrichments to a similar degree as the crystallographic waters, which makes this strategy applicable to structures without experimental knowledge of water positions. Finally, we used decision trees to select an ensemble of structures with different water molecule positions and orientations that outperforms any single structure with water molecules. The approach presented here is validated against independent test sets of A2A receptor antagonists and decoys from the literature. In general, this water optimization strategy could be applied to any target with waters-mediated protein-ligand interactions.

DrugEx v3: scaffold-constrained drug design with graph transformer-based reinforcement learning.

Rational drug design often starts from specific scaffolds to which side chains/substituents are added or modified due to the large drug-like chemical space available to search for novel drug-like molecules. With the rapid growth of deep learning in drug discovery, a variety of effective approaches have been developed for de novo drug design. In previous work we proposed a method named DrugEx, which can be applied in polypharmacology based on multi-objective deep reinforcement learning. However, the previous version is trained under fixed objectives and does not allow users to input any prior information (i.e. a desired scaffold). In order to improve the general applicability, we updated DrugEx to design drug molecules based on scaffolds which consist of multiple fragments provided by users. Here, a  Transformer model was employed to generate molecular structures. The Transformer is a multi-head self-attention deep learning model containing an encoder to receive scaffolds as input and a decoder to generate molecules as output. In order to deal with the graph representation of molecules a novel positional encoding for each atom and bond based on an adjacency matrix was proposed, extending the architecture of the Transformer. The graph Transformer model contains growing and connecting procedures for molecule generation starting from  a given scaffold based on fragments. Moreover, the generator was trained under a reinforcement learning framework to increase the number of desired ligands. As a proof of concept, the method was applied to design ligands for the adenosine A2A receptor (A2AAR) and compared with SMILES-based methods. The results show that 100% of the generated molecules are valid and most of them had a high predicted affinity value towards A2AAR with given scaffolds.

DrugEx v2: de novo design of drug molecules by Pareto-based multi-objective reinforcement learning in polypharmacology.

In polypharmacology drugs are required to bind to multiple specific targets, for example to enhance efficacy or to reduce resistance formation. Although deep learning has achieved a breakthrough in de novo design in drug discovery, most of its applications only focus on a single drug target to generate drug-like active molecules. However, in reality drug molecules often interact with more than one target which can have desired (polypharmacology) or undesired (toxicity) effects. In a previous study we proposed a new method named DrugEx that integrates an exploration strategy into RNN-based reinforcement learning to improve the diversity of the generated molecules. Here, we extended our DrugEx algorithm with multi-objective optimization to generate drug-like molecules towards multiple targets or one specific target while avoiding off-targets (the two adenosine receptors, A1AR and A2AAR, and the potassium ion channel hERG in this study). In our model, we applied an RNN as the agent and machine learning predictors as the environment. Both the agent and the environment were pre-trained in advance and then interplayed under a reinforcement learning framework. The concept of evolutionary algorithms was merged into our method such that crossover and mutation operations were implemented by the same deep learning model as the agent. During the training loop, the agent generates a batch of SMILES-based molecules. Subsequently scores for all objectives provided by the environment are used to construct Pareto ranks of the generated molecules. For this ranking a non-dominated sorting algorithm and a Tanimoto-based crowding distance algorithm using chemical fingerprints are applied. Here, we adopted GPU acceleration to speed up the process of Pareto optimization. The final reward of each molecule is calculated based on the Pareto ranking with the ranking selection algorithm. The agent is trained under the guidance of the reward to make sure it can generate desired molecules after convergence of the training process. All in all we demonstrate generation of compounds with a diverse predicted selectivity profile towards multiple targets, offering the potential of high efficacy and low toxicity.

A novel chemogenomics analysis of G protein-coupled receptors (GPCRs) and their ligands: a potential strategy for receptor de-orphanization.

BACKGROUND: G protein-coupled receptors (GPCRs) represent a family of well-characterized drug targets with significant therapeutic value. Phylogenetic classifications may help to understand the characteristics of individual GPCRs and their subtypes. Previous phylogenetic classifications were all based on the sequences of receptors, adding only minor information about the ligand binding properties of the receptors. In this work, we compare a sequence-based classification of receptors to a ligand-based classification of the same group of receptors, and evaluate the potential to use sequence relatedness as a predictor for ligand interactions thus aiding the quest for ligands of orphan receptors. RESULTS: We present a classification of GPCRs that is purely based on their ligands, complementing sequence-based phylogenetic classifications of these receptors. Targets were hierarchically classified into phylogenetic trees, for both sequence space and ligand (substructure) space. The overall organization of the sequence-based tree and substructure-based tree was similar; in particular, the adenosine receptors cluster together as well as most peptide receptor subtypes (e.g. opioid, somatostatin) and adrenoceptor subtypes. In ligand space, the prostanoid and cannabinoid receptors are more distant from the other targets, whereas the tachykinin receptors, the oxytocin receptor, and serotonin receptors are closer to the other targets, which is indicative for ligand promiscuity. In 93% of the receptors studied, de-orphanization of a simulated orphan receptor using the ligands of related receptors performed better than random (AUC > 0.5) and for 35% of receptors de-orphanization performance was good (AUC > 0.7). CONCLUSIONS: We constructed a phylogenetic classification of GPCRs that is solely based on the ligands of these receptors. The similarities and differences with traditional sequence-based classifications were investigated: our ligand-based classification uncovers relationships among GPCRs that are not apparent from the sequence-based classification. This will shed light on potential cross-reactivity of GPCR ligands and will aid the design of new ligands with the desired activity profiles. In addition, we linked the ligand-based classification with a ligand-focused sequence-based classification described in literature and proved the potential of this method for de-orphanization of GPCRs.

Quantitative prediction of selectivity between the A1 and A2A adenosine receptors.

The development of drugs is often hampered due to off-target interactions leading to adverse effects. Therefore, computational methods to assess the selectivity of ligands are of high interest. Currently, selectivity is often deduced from bioactivity predictions of a ligand for multiple targets (individual machine learning models). Here we show that modeling selectivity directly, by using the affinity difference between two drug targets as output value, leads to more accurate selectivity predictions. We test multiple approaches on a dataset consisting of ligands for the A1 and A2A adenosine receptors (among others classification, regression, and we define different selectivity classes). Finally, we present a regression model that predicts selectivity between these two drug targets by directly training on the difference in bioactivity, modeling the selectivity-window. The quality of this model was good as shown by the performances for fivefold cross-validation: ROC A1AR-selective 0.88 ± 0.04 and ROC A2AAR-selective 0.80 ± 0.07. To increase the accuracy of this selectivity model even further, inactive compounds were identified and removed prior to selectivity prediction by a combination of statistical models and structure-based docking. As a result, selectivity between the A1 and A2A adenosine receptors was predicted effectively using the selectivity-window model. The approach presented here can be readily applied to other selectivity cases.

An exploration strategy improves the diversity of de novo ligands using deep reinforcement learning: a case for the adenosine A2A receptor.

Over the last 5 years deep learning has progressed tremendously in both image recognition and natural language processing. Now it is increasingly applied to other data rich fields. In drug discovery, recurrent neural networks (RNNs) have been shown to be an effective method to generate novel chemical structures in the form of SMILES. However, ligands generated by current methods have so far provided relatively low diversity and do not fully cover the whole chemical space occupied by known ligands. Here, we propose a new method (DrugEx) to discover de novo drug-like molecules. DrugEx is an RNN model (generator) trained through reinforcement learning which was integrated with a special exploration strategy. As a case study we applied our method to design ligands against the adenosine A2A receptor. From ChEMBL data, a machine learning model (predictor) was created to predict whether generated molecules are active or not. Based on this predictor as the reward function, the generator was trained by reinforcement learning without any further data. We then compared the performance of our method with two previously published methods, REINVENT and ORGANIC. We found that candidate molecules our model designed, and predicted to be active, had a larger chemical diversity and better covered the chemical space of known ligands compared to the state-of-the-art.