Kinome-Wide Virtual Screening by Multi-Task Deep Learning.

Deep learning is a machine learning technique to model high-level abstractions in data by utilizing a graph composed of multiple processing layers that experience various linear and non-linear transformations. This technique has been shown to perform well for applications in drug discovery, utilizing structural features of small molecules to predict activity. Here, we report a large-scale study to predict the activity of small molecules across the human kinome-a major family of drug targets, particularly in anti-cancer agents. While small-molecule kinase inhibitors exhibit impressive clinical efficacy in several different diseases, resistance often arises through adaptive kinome reprogramming or subpopulation diversity. Polypharmacology and combination therapies offer potential therapeutic strategies for patients with resistant diseases. Their development would benefit from a more comprehensive and dense knowledge of small-molecule inhibition across the human kinome. Leveraging over 650,000 bioactivity annotations for more than 300,000 small molecules, we evaluated multiple machine learning methods to predict the small-molecule inhibition of 342 kinases across the human kinome. Our results demonstrated that multi-task deep neural networks outperformed classical single-task methods, offering the potential for conducting large-scale virtual screening, predicting activity profiles, and bridging the gaps in the available data.

Large-scale comparison of machine learning methods for profiling prediction of kinase inhibitors.

Conventional machine learning (ML) and deep learning (DL) play a key role in the selectivity prediction of kinase inhibitors. A number of models based on available datasets can be used to predict the kinase profile of compounds, but there is still controversy about the advantages and disadvantages of ML and DL for such tasks. In this study, we constructed a comprehensive benchmark dataset of kinase inhibitors, involving in 141,086 unique compounds and 216,823 well-defined bioassay data points for 354 kinases. We then systematically compared the performance of 12 ML and DL methods on the kinase profiling prediction task. Extensive experimental results reveal that (1) Descriptor-based ML models generally slightly outperform fingerprint-based ML models in terms of predictive performance. RF as an ensemble learning approach displays the overall best predictive performance. (2) Single-task graph-based DL models are generally inferior to conventional descriptor- and fingerprint-based ML models, however, the corresponding multi-task models generally improves the average accuracy of kinase profile prediction. For example, the multi-task FP-GNN model outperforms the conventional descriptor- and fingerprint-based ML models with an average AUC of 0.807. (3) Fusion models based on voting and stacking methods can further improve the performance of the kinase profiling prediction task, specifically, RF::AtomPairs + FP2 + RDKitDes fusion model performs best with the highest average AUC value of 0.825 on the test sets. These findings provide useful information for guiding choices of the ML and DL methods for the kinase profiling prediction tasks. Finally, an online platform called KIPP ( https://kipp.idruglab.cn ) and python software are developed based on the best models to support the kinase profiling prediction, as well as various kinase inhibitor identification tasks including virtual screening, compound repositioning and target fishing.

Pyrrolopyrimidine based CSF1R inhibitors: Attempted departure from Flatland.

The colony-stimulating factor 1 receptor (CSF1R) is an attractive target for inflammation disorders and cancers. Based on a series of pyrrolo[2,3-d]pyrimidine containing two carbo-aromatic rings, we have searched for new CSF1R inhibitors having a higher fraction of sp3-atoms. The phenyl unit in the 4-amino group could efficiently be replaced by tetrahydropyran (THP) retaining inhibitor potency. Exchanging the 6-aryl group with cyclohex-2-ene units also resulted in highly potent compounds, while fully saturated ring systems at C-6 led to a loss of activity. The structure-activity relationship study evaluating THP containing pyrrolo[2,3-d]pyrimidine derivates identified several highly active inhibitors by enzymatic studies. A comparison of 11 pairs of THP and aromatic compounds showed that inhibitors containing THP had clear benefits in terms of enzymatic potency, solubility, and cell toxicity. Guided by cellular experiments in Ba/F3 cells, five CSF1R inhibitors were further profiled in ADME assays, indicating the para-aniline derivative 16t as the most attractive compound for further development.

Artificial intelligence methods in kinase target profiling: Advances and challenges.

Kinases have a crucial role in regulating almost the full range of cellular processes, making them essential targets for therapeutic interventions against various diseases. Accurate kinase-profiling prediction is vital for addressing the selectivity/specificity challenges in kinase drug discovery, which is closely related to lead optimization, drug repurposing, and the understanding of potential drug side effects. In this review, we provide an overview of the latest advancements in machine learning (ML)-based and deep learning (DL)-based quantitative structure-activity relationship (QSAR) models for kinase profiling. We highlight current trends in this rapidly evolving field and discuss the existing challenges and future directions regarding experimental data set construction and model architecture design. Our aim is to offer practical insights and guidance for the development and utilization of these approaches.

NanoBRET™ Live-Cell Kinase Selectivity Profiling Adapted for High-Throughput Screening.

Kinases represent one of the most therapeutically tractable targets for drug discovery in the twenty-first century. However, confirming engagement and achieving intracellular kinase selectivity for small-molecule kinase inhibitors can represent noteworthy challenges. The NanoBRETTM platform enables broad-spectrum live-cell kinase selectivity profiling in most laboratory settings, without advanced instrumentation or expertise. However, the prototype workflow for this selectivity profiling is currently limited to manual liquid handling and 96-well plates. Herein, we describe a scalable workflow with automation and acoustic dispensing, thus dramatically improving the throughput. Such adaptations enable profiling of larger compound sets against 192 full-length protein kinases in live cells, with statistical robustness supporting quantitative analysis.

Design, synthesis, and antitumor efficacy of novel 5-deazaflavin derivatives backed by kinase screening, docking, and ADME studies.

Novel 5-deazaflavins were designed as potential anticancer candidates. Compounds 4j, 4k, 5b, 5i, and 9f demonstrated high cytotoxicity against MCF-7 cell line with IC50 of 0.5-190nM. Compounds 8c and 9g showed preferential activity against Hela cells (IC50: 1.69 and 1.52 µM respectively). However, compound 5d showed notable potency against MCF-7 and Hela cell lines of 0.1 nM and 1.26 µM respectively. Kinase profiling for 4e showed the highest inhibition against a 20 kinase panel. Additionally, ADME prediction studies exhibited that compounds 4j, 5d, 5f, and 9f have drug-likeness criteria to be considered promising antitumor agents deserving of further investigation. SAR study showed that substitutions with 2-benzylidene hydra zino have a better fitting into PTK with enhanced antiproliferative potency. Noteworthy, the incorporation of hydrazino or ethanolamine moieties at position 2 along with small alkyl or phenyl at N-10, respectively revealed an extraordinary potency against MCF-7 cells with IC50 values in the nanomolar range.

1,3-diarylpyrazolones as potential anticancer agents for non-small cell lung cancer: Synthesis and antiproliferative activity evaluation.

A series of pyrazolone compounds with different substitution patterns have been synthesized using microwave-assisted methods and evaluated their in vitro antiproliferative activity against human lung adenocarcinoma cell lines (A549 and NCI-H522). Among the tested compounds, the pyrazolone P7 exhibited high antiproliferative activity against both A549 and NCIH522 cancer cell lines while being 10 times less cytotoxic to non-cancerous cells. Moreover, our compounds P7 and P11 exhibited higher antiproliferative activity and selectivity against A549 and NCIH522 cells compared with the clinically approved drugs Afatinib and Gefitinib. The cell cycle analysis showed that the compound P7 and P11 arrests the cell cycle at G0/G1 phase, whereas the compounds P13 and P14 involved in G2/M phase arrest. The results from antiproliferative activity screening, cell cycle analysis, and kinase profiling indicate that the suitably substituted 1,3-diarylpyrazolones exhibit high antiproliferative activity against non-small cell lung cancer cells.

ProfKin: A comprehensive web server for structure-based kinase profiling.

Protein kinases are central mediators of signal-transduction cascades and attractive drug targets for therapeutic intervention. Since kinases are structurally and mechanistically related to each other, kinase inhibitor selectivity is often investigated by kinase profiling and considered as an important index for drug discovery. We here describe a versatile web server termed ProfKin for structure-based kinase profiling, which is based on a kinase-ligand focused database (KinLigDB). It provides all ready-to-use 3D structure coordinates of 4219 kinase-ligand complex structures covering 297 human kinases and the associated information, particularly including binding site type, binding ligand type, interaction fingerprints, downstream molecules and related human diseases. The web server works via predicting possible binding modes for the query molecule, prioritizing the binding modes guided by an interaction fingerprint analysis method, and giving a list of ranked kinases by a comprehensive index. Users can freely select entire or part of the KinLigDB database, e.g. via subfamily and binding site type, to customize the profiling contents. The superimpositions of the predicted binding poses of the query molecule with reference binding modes can be visually inspected on the website. The additional classification attributes and phylogenetic tree are also given for each top-ranked kinase.

Identification of 3, 4-disubstituted pyridine derivatives as novel CDK8 inhibitors.

Selective inhibition of cyclin-dependent kinase 8 (CDK8) has been recently regarded as a potential approach for cancer therapy. A series of novel CDK8 inhibitors with the pyridine core was identified via scaffold hopping from the known CDK8 inhibitor A-7. The new inhibitors were designed to improve the ligand efficiency so as to enhance drug-likeness. Most of the compounds showed significant inhibition against CDK8/cyclin C, and the most active compounds (5d, 5e and 7') displayed IC50 values of 2.4 nM, 5.0 nM and 7.7 nM, respectively. Preliminary kinase profiling of selected compounds against a panel of kinases from different families indicated that this compound class might selectively inhibit CDK8 as well as its paralog CDK19. Some compounds exhibited cellular activity in both MTT and SRB assays against a variety of tumor cells, including HCT-116, A549, MDA-MB-231, KB, KB-VIN and MCF-7. Further flow cytometry analysis revealed a dose-dependent G2/M phase arrest in MDA-MB-231 cells treated with compounds 6'a, 6'b, 6'j and 6'k. In addition, compound 6'k demonstrated moderate antitumor efficacy in HCT-116 mouse models, although unfavorable pharmacokinetic profiles were suggested by preliminary study in mice. The results provided a new structural prototype for the search of selective CDK8 inhibitors as antitumor agents.

The retinal tyrosine kinome of diabetic Akimba mice highlights potential for specific Src family kinase inhibition in retinal vascular disease.

Although anti-VEGF therapies have radically changed clinical practice, there is still an urgent demand for novel, integrative approaches for sight-threatening retinal vascular diseases. As we hypothesize that protein tyrosine kinases are key signaling mediators in retinal vascular disease, we performed a comprehensive activity-based tyrosine kinome profiling on retinal tissue of 12-week-old Akimba mice, a translational model displaying hallmarks of early and advanced diabetic retinopathy. Western blotting was used to confirm retinal tyrosine kinase activity in Akimba mice. HUVEC tube formation and murine organotypic choroidal sprouting assays were applied to compare tyrosine kinase inhibitors with different specificity profiles. HUVEC toxicity and proliferation were evaluated using the CellTox™ Green Cytotoxicity and PrestoBlue™ Assays. Our results indicate a shift of the Akimba retinal tyrosine kinome towards a hyperactive state. Functional network analysis of significantly hyperphosphorylated peptides and upstream kinase prediction revealed a central role for Src-FAK family kinases. Western blotting confirmed hyperactivity of this signaling node in the retina of Akimba mice. We demonstrated that not only Src but also FAK family kinase inhibitors with different selectivity profiles were able to suppress angiogenesis in vitro and ex vivo. In the latter model, the novel selective Src family kinase inhibitor eCF506 was able to achieve potent reduction of angiogenesis, comparable to the less specific inhibitor Dasatinib. None of the tested compounds demonstrated acute endothelial cell toxicity. Overall, the collected findings provide the first comprehensive overview of retinal tyrosine kinome changes in the Akimba model of diabetic retinopathy and for the first time highlight Src family kinase inhibition using highly specific inhibitors as an attractive therapeutic intervention for retinal vascular pathology.

Promiscuity analysis of a kinase panel screen with designated p38 alpha inhibitors.

Protein phosphorylation by kinases is of critical importance for the regulation of many cellular functions. When kinases are deregulated numerous biological processes are affected, which may cause a variety of diseases. Therefore, kinase inhibition plays an important role for therapeutic intervention. A number of kinase inhibitors have been approved as drugs, initially in oncology where promiscuous (multi-kinase) inhibitors were most efficacious. Exploring kinase inhibitor selectivity and promiscuity for therapy is among the most challenging aspects of kinase drug discovery. Herein, we thoroughly analyze a kinase profiling experiment in which 637 designated inhibitors of p38α MAP kinase (p38α) were tested against a panel of 60 kinases distributed across the human kinome. In this experiment, only 19% of the inhibitors were found to be promiscuous when the median p38α inhibition level was applied as an activity threshold. Promiscuous inhibitors had a median value of two targets per compound, and many of these inhibitors were only active against the p38α and closely related JNK3 enzymes. Promiscuity cliffs were identified and analyzed in a network representation revealing structural modifications that were implicated in triggering compound promiscuity. Taken together, the findings revealed a high degree of selectivity of designated p38α directed inhibitors although they target the ATP binding site that is largely conserved across the human kinome.

Constitutively active GSK3ß as a means to bolster dendritic cell functionality in the face of tumour-mediated immune suppression.

In patients with cancer, the functionality of Dendritic Cells (DC) is hampered by high levels of tumor-derived suppressive cytokines, which interfere with DC development and maturation. Poor DC development can limit the efficacy of immune checkpoint blockade and in vivo vaccination approaches. Interference in intracellular signaling cascades downstream from the receptors of major tumor-associated suppressive cytokines like IL-10 and IL-6, might improve DC development and activation, and thus enhance immunotherapy efficacy. We performed exploratory functional screens on arrays consisting of >1000 human kinase peptide substrates to identify pathways involved in DC development and its inhibition by IL-10 or IL-6. The resulting alterations in phosphorylation of the kinome substrate profile pointed to glycogen-synthase kinase-3ß (GSK3ß) as a pivotal kinase in both DC development and suppression. GSK3ß inhibition blocked human DC differentiation in vitro, which was accompanied by decreased levels of IL-12p70 secretion, and a reduced capacity for T cell priming. More importantly, adenoviral transduction of monocytes with a constitutively active form of GSK3ß induced resistance to the suppressive effects of IL-10 and melanoma-derived supernatants alike, resulting in improved DC development, accompanied by up-regulation of co-stimulatory markers, an increase in CD83 expression levels in mature DC, and diminished release of IL-10. Moreover, adenovirus-mediated intratumoral manipulation of this pathway in an in vivo melanoma model resulted in DC activation and recruitment, and in improved immune surveillance and tumor control. We propose the induction of constitutive GSK3ß activity as a novel therapeutic means to bolster DC functionality in the tumor microenvironment.

A Flexible Workflow for Automated Bioluminescent Kinase Selectivity Profiling.

Kinase profiling during drug discovery is a necessary process to confirm inhibitor selectivity and assess off-target activities. However, cost and logistical limitations prevent profiling activities from being performed in-house. We describe the development of an automated and flexible kinase profiling workflow that combines ready-to-use kinase enzymes and substrates in convenient eight-tube strips, a bench-top liquid handling device, ADP-Glo Kinase Assay (Promega, Madison, WI) technology to quantify enzyme activity, and a multimode detection instrument. Automated methods were developed for kinase reactions and quantification reactions to be assembled on a Gilson (Middleton, WI) PIPETMAX, following standardized plate layouts for single- and multidose compound profiling. Pipetting protocols were customized at runtime based on user-provided information, including compound number, increment for compound titrations, and number of kinase families to use. After the automated liquid handling procedures, a GloMax Discover (Promega) microplate reader preloaded with SMART protocols was used for luminescence detection and automatic data analysis. The functionality of the automated workflow was evaluated with several compound-kinase combinations in single-dose or dose-response profiling formats. Known target-specific inhibitions were confirmed. Novel small molecule-kinase interactions, including off-target inhibitions, were identified and confirmed in secondary studies. By adopting this streamlined profiling process, researchers can quickly and efficiently profile compounds of interest on site.

Novel optimization of valmerins (tetrahydropyrido[1,2-a]isoindolones) as potent dual CDK5/GSK3 inhibitors.

An efficient synthetic strategy able to modulate the structure of the tetrahydropyridine isoindolone (Valmerin) skeleton was developed. A library of more than 30 novel final structures was generated. Biological activities on CDK5 and GSK3 as well as cellular effects on cancer cell lines were measured for each novel compound. Additionally to support the SAR, a docking study was performed. A potent GSK3/CDK5 dual inhibitor (37, IC50 CDK5/GSK3 35/7 nM) was obtained. Best antiproliferative effects were obtained on lung and prostate cell lines with IC50 = 20 nM.

Discovery of Novel Antiangiogenic Marine Natural Product Scaffolds.

Marine natural products (MNPs) are recognized for their structural complexity, diversity, and novelty. The vast majority of MNPs are pharmacologically relevant through their ability to modulate macromolecular targets underlying human diseases. Angiogenesis is a fundamental process in cancer progression and metastasis. Targeting angiogenesis through selective modulation of linked protein kinases is a valid strategy to discover novel effective tumor growth and metastasis inhibitors. An in-house marine natural products mini-library, which comprises diverse MNP entities, was submitted to the Lilly's Open Innovation Drug Discovery platform. Accepted structures were subjected to in vitro screening to discover mechanistically novel angiogenesis inhibitors. Active hits were subjected to additional angiogenesis-targeted kinase profiling. Some natural and semisynthetic MNPs, including multiple members of the macrolide latrunculins, the macrocyclic oxaquinolizidine alkaloid araguspongine C, and the sesquiterpene quinone puupehenone, showed promising results in primary and secondary angiogenesis screening modules. These hits inhibited vascular endothelial growth factor (VEGF)-mediated endothelial tube-like formation, with minimal cytotoxicity at relevant doses. Secondary kinase profiling identified six target protein kinases, all involved in angiogenesis signaling pathways. Molecular modeling and docking experiments aided the understanding of molecular binding interactions, identification of pharmacophoric epitopes, and deriving structure-activity relationships of active hits. Marine natural products are prolific resources for the discovery of chemically and mechanistically unique selective antiangiogenic scaffolds.

Using Bioluminescent Kinase Profiling Strips to Identify Kinase Inhibitor Selectivity and Promiscuity.

The advancement of a kinase inhibitor throughout drug discovery and development is predicated upon its selectivity towards the target of interest. Thus, profiling the compound against a broad panel of kinases is important for providing a better understanding of its activity and for obviating any off-target activities that can result in undesirable consequences. To assess the selectivity and potency of an inhibitor against multiple kinases, it is desirable to use a universal assay that can monitor the activity of all classes of kinases regardless of the nature of their substrates. The luminescent ADP-Glo kinase assay is a universal platform that measures kinase activity by quantifying the amount of the common kinase reaction product ADP. Here we present a method using standardized kinase profiling systems for inhibitor profiling studies based on ADP detection by luminescence. The kinase profiling systems are sets of kinases organized by family, presented in multi-tube strips containing eight enzymes, each with corresponding substrate strips, and standardized for optimal kinase activity. We show that using the kinase profiling strips we could quickly and easily generate multiple selectivity profiles using small or large kinase panels, and identify compound promiscuity within the kinome.

Bioluminescent kinase strips: A novel approach to targeted and flexible kinase inhibitor profiling.

In addition to target efficacy, drug safety is a major requirement during the drug discovery process and is influenced by target specificity. Therefore, it is imperative that every new drug candidate be profiled against various liability panels that include protein kinases. Here, an effective methodology to streamline kinase inhibitor profiling is described. An accessible standardized profiling system for 112 protein kinases covering all branches of the kinome was developed. This approach consists of creating different sets of kinases and their corresponding substrates in multi-tube strips. The kinase stocks are pre-standardized for optimal kinase activity and used for inhibitor profiling using a bioluminescent ADP detection assay. We show that these strips can routinely generate inhibitor selectivity profiles for small or broad kinase family panels. Lipid kinases were also assembled in strip format and profiled together with protein kinases. We identified two specific PI3K inhibitors that have off-target effects on CK2 that were not reported before and would have been missed if compounds were not profiled against lipid and protein kinases simultaneously. To validate the accuracy of the data generated by this method, we confirmed that the inhibition potencies observed are consistent with published values produced by more complex technologies such as radioactivity assays.

Examining the influence of specificity ligands and ATP-competitive ligands on the overall effectiveness of bivalent kinase inhibitors.

BACKGROUND: Identifying selective kinase inhibitors remains a major challenge. The design of bivalent inhibitors provides a rational strategy for accessing potent and selective inhibitors. While bivalent kinase inhibitors have been successfully designed, no comprehensive assessment of affinity and selectivity for a series of bivalent inhibitors has been performed. Here, we present an evaluation of the structure activity relationship for bivalent kinase inhibitors targeting ABL1. METHODS: Various SNAPtag constructs bearing different specificity ligands were expressed in vitro. Bivalent inhibitor formation was accomplished by synthesizing individual ATP-competitive kinase inhibitors containing a SNAPtag targeting moiety, enabling the spontaneous self-assembly of the bivalent inhibitor. Assembled bivalent inhibitors were incubated with K562 lysates, and then subjected to affinity enrichment using various ATP-competitive inhibitors immobilized to sepharose beads. Resulting eluents were analyzed using Tandem Mass Tag (TMT) labeling and two-dimensional liquid chromatography-tandem mass spectrometry (2D-LC-MS/MS). Relative binding affinity of the bivalent inhibitor was determined by calculating the concentration at which 50% of a given kinase remained bound to the affinity matrix. RESULTS: The profiling of three parental ATP-competitive inhibitors and nine SNAPtag conjugates led to the identification of 349 kinase proteins. In all cases, the bivalent inhibitors exhibited enhanced binding affinity and selectivity for ABL1 when compared to the parental compound conjugated to SNAPtag alone. While the rank order of binding affinity could be predicted by considering the binding affinities of the individual specificity ligands, the resulting affinity of the assembled bivalent inhibitor was not predictable. The results from this study suggest that as the potency of the ATP-competitive ligand increases, the contribution of the specificity ligand towards the overall binding affinity of the bivalent inhibitor decreases. However, the affinity of the specificity components in its interaction with the target is essential for achieving selectivity. CONCLUSION: Through comprehensive chemical proteomic profiling, this work provides the first insight into the influence of ATP-competitive and specificity ligands binding to their intended target on a proteome-wide scale. The resulting data suggest a subtle interplay between the ATP-competitive and specificity ligands that cannot be accounted for by considering the specificity or affinity of the individual components alone.

7,7'-Diazaindirubin--a small molecule inhibitor of casein kinase 2 in vitro and in cells.

Aza- and diaza-bisindoles were synthesized by coupling of 7-azaisatin, 7-azaoxindol, 7-azaindoxyl acetate, and their non-aza counterparts, respectively. Whereas 7,7'-diazaindigo (10) and 7,7'-diazaisoindigo (11) did not show antiproliferative activity in several human tumor cell lines up to 100 µM, 7-azaindirubin (12) and 7'-azaindirubin (13) were more active than the parent molecule, indirubin, in LXFL529L cells (human large cell lung tumor xenograft), and 7,7'-diazaindirubin (14) was exhibiting substantially enhanced growth inhibitory activity in these cells. In the NCI 60 cell line panel, 14 displayed antiproliferative activity preferentially in certain melanoma and non-small cell lung cancer cells. In contrast to the potent serine/threonine/tyrosine kinase inhibition observed for indirubins, kinase inhibition profiling of 14 in 220 kinases revealed largely a loss of kinase inhibitory activity towards most kinases, with retained inhibitory activity for just a few kinases. At 1 µM concentration, especially casein kinases CK1γ3, CK2α, CK2α2, and SIK were inhibited by more than 50%. In cell-based assays, 14 markedly affected CK2-mediated signaling in various human tumor cells. In MCF7 cells, 14 induced cell cycle arrest at G1 and G2/M and apoptosis, whereas CK2-deficient MCF7 cells were resistant. These findings reveal a novel key mechanism of action for 14, suggesting primarily CK2 inhibition to be causally related to growth inhibition of human tumor cells.

Hit-to-lead optimization and kinase selectivity of imidazo[1,2-a]quinoxalin-4-amine derived JNK1 inhibitors.

As the result of a rhJNK1 HTS, the imidazo[1,2-a]quinoxaline 1 was identified as a 1.6 µM rhJNK1 inhibitor. Optimization of this compound lead to AX13587 (rhJNK1 IC50=160 nM) which was co-crystallized with JNK1 to identify key molecular interactions. Kinase profiling against 125+ kinases revealed AX13587 was an inhibitor of JNK, MAST3, and MAST4 whereas its methylene homolog AX14373 (native JNK1 IC50=47 nM) was a highly specific JNK inhibitor.