Finding new scaffolds of JAK3 inhibitors in public database: 3D-QSAR models & shape-based screening

Positioning of an unprecedented spiro[5.5]undeca ring system into kinase inhibitor space

In-house 1,5-oxaza spiroquinone 1, with spiro[5.5]undeca ring system, was announced as an unprecedented anti-inflammatory scaffold through chemistry-oriented synthesis (ChOS), a chemocentric approach. Herein, we studied how to best position the spiro[5.5]undeca ring system in kinase inhibitor space. Notably, late-stage modification of the scaffold 1 into compounds 2a-r enhanced kinase-likeness of the scaffold 1. The improvement could be depicted with (1) selectivity with target shift (from JNK-1 into GSK-3) and (2) potency (> 20-fold). In addition, ATP independent IC₅₀ of compound 2j suggested a unique binding mode of this scaffold between ATP site and substrate site, which was explained by docking based optimal site selection and molecular dynamic simulations of the optimal binding site. Despite the shift of kinase profiling, the anti-inflammatory activity of compounds 2a-r could be retained in hyperactivated microglial cells.

Repurposing of FDA approved ring systems through bi-directional target-ring system dual screening

In drug repurposing approaches, the chemically diverse and potentially safe molecules can be explored as therapeutic potential beyond those originally targeted indications. However, accessible information on a limited number of drug pipelines can lead to competitive over-heating issues, and intellectual property rights also restrict the free investigation in chemical space. As a complementary approach to the drawbacks, ring systems of approved drugs (instead of clinical drugs) can be optimized and used for repurposing purposes. In this study, bi-directional target (T) and ring system (R) dual screening (TR screening) was developed for the repurposing of their rarely used ring systems from FDA approved drugs. The TR screening suggested RAR β and cyproheptadine as the best pair of target and ring system to escape a saddle point. The selected ring system was virtually grown and elaborated with the defined criteria: synthesizability, drug-likeness, and docking pose showing the top scores. The achieved compounds were synthesized and biologically tested with an acceptable ADME/T profile.

Comparing a Query Compound with Drug Target Classes Using 3D-Chemical Similarity

3D similarity is useful in predicting the profiles of unprecedented molecular frameworks that are 2D dissimilar to known compounds. When comparing pairs of compounds, 3D similarity of the pairs depends on conformational sampling, the alignment method, the chosen descriptors, and the similarity coefficients. In addition to these four factors, 3D chemocentric target prediction of an unknown compound requires compound-target associations, which replace compound-to-compound comparisons with compound-to-target comparisons. In this study, quantitative comparison of query compounds to target classes (one-to-group) was achieved via two types of 3D similarity distributions for the respective target class with parameter optimization for the fitting models: (1) maximum likelihood (ML) estimation of queries, and (2) the Gaussian mixture model (GMM) of target classes. While Jaccard-Tanimoto similarity of query-to-ligand pairs with 3D structures (sampled multi-conformers) can be transformed into query distribution using ML estimation, the ligand pair similarity within each target class can be transformed into a representative distribution of a target class through GMM, which is hyperparameterized via the expectation-maximization (EM) algorithm. To quantify the discriminativeness of a query ligand against target classes, the Kullback-Leibler (K-L) divergence of each query was calculated and compared between targets. 3D similarity-based K-L divergence together with the probability and the feasibility index, (F_m), showed discriminative power with regard to some query-class associations. The K-L divergence of 3D similarity distributions can be an additional method for (1) the rank of the 3D similarity score or (2) the p-value of one 3D similarity distribution to predict the target of unprecedented drug scaffolds.

Evaluation of 3-Dimensionality in Approved and Experimental Drug Space

The 3-dimensional (3D) structure of therapeutics and other bioactive molecules is an important factor in determining the strength and selectivity of their protein-ligand interactions. Previous efforts have considered the strain introduced and tolerated through conformational changes induced upon protein binding. Herein, we present an analysis of 3-dimentionality for energy-minimized structures from the DrugBank and ligands bound to proteins identified in the Protein Data Bank (PDB). This analysis reveals that the majority of molecules found in both the DrugBank and the PDB tend toward linearity and planarity, with few molecules having highly 3D conformations. Decidedly 3D geometries have been historically difficult to achieve, likely due to the synthetic challenge of making 3D organic molecules, and other considerations, such as adherence to the 'rule-of-five'. This has resulted in the dominance of planar and/or linear topologies of the molecules described here. Strategies to address the generally flat nature of these data sets are explored, including the use of 3D organic fragments and inorganic scaffolds as a means of accessing privileged 3D space. This work highlights the potential utility of libraries with greater 3D topological diversity so that the importance of molecular shape to biological behavior can be more fully understood in drug discovery campaigns.

Development of Predictive Models for Identifying Potential S100A9 Inhibitors Based on Machine Learning Methods

S100A9 is a potential therapeutic target for various disease including prostate cancer, colorectal cancer, and Alzheimer's disease. However, the sparsity of atomic level data, such as protein-protein interaction of S100A9 with RAGE, TLR4/MD2, or CD147 (EMMPRIN) hinders the rational drug design of S100A9 inhibitors. Herein we first report predictive models of S100A9 inhibitory effect by applying machine learning classifiers on 2D-molecular descriptors. The models were optimized through feature selectors as well as classifiers to produce the top eight random forest models with robust predictability and high cost-effectiveness. Notably, optimal feature sets were obtained after the reduction of 2,798 features into dozens of features with the chopping of fingerprint bits. Moreover, the high efficiency of compact feature sets allowed us to further screen a large-scale dataset (over 6,000,000 compounds) within a week. Through a consensus vote of the top models, 46 hits (hit rate = 0.000713%) were identified as potential S100A9 inhibitors. We expect that our models will facilitate the drug discovery process by providing high predictive power as well as cost-reduction ability and give insights into designing novel drugs targeting S100A9.

Computational Simulations Identified Two Candidate Inhibitors of Cdk5/p25 to Abrogate Tau-associated Neurological Disorders

Deregulation of Cdk5 is a hallmark in neurodegenerative diseases and its complex with p25 forms Cdk5/p25, thereby causes severe neuropathological insults. Cdk5/p25 abnormally phosphorylates tau protein, and induces tau-associated neurofibrillary tangles in neurological disorders. Therefore, the pharmacological inhibition of Cdk5/p25 alleviates tau-associated neurological disorders. Herein, computational simulations probed two candidate inhibitors of Cdk5/p25. Structure-based pharmacophore investigated the essential complementary chemical features of ATP-binding site of Cdk5 in complex with roscovitine. Resultant pharmacophore harbored polar interactions with Cys83 and Asp86 residues and non-polar interactions with Ile10, Phe80, and Lys133 residues of Cdk5. The chemical space of selected pharmacophore was comprised of two hydrogen bond donors, one hydrogen bond acceptor, and three hydrophobic features. Decoy test validation of pharmacophore obtained highest Guner-Henry score (0.88) and enrichment factor score (7.23). The screening of natural product drug-like databases by validated pharmacophore retrieved 1126 compounds as candidate inhibitors of Cdk5/p25. The docking of candidate inhibitors filtered 10 molecules with docking score >80.00 and established polar and non-polar interactions with the ATP-binding site residues of Cdk5/p25. Finally, molecular dynamics simulation and binding free energy analyses identified two candidate inhibitors of Cdk5/p25. During 30 ns simulation, the candidate inhibitors established <3.0 Å root mean square deviation and stable hydrogen bond interactions with the ATP-binding site residues of Cdk5/p25. The final candidate inhibitors obtained lowest binding free energies of -122.18 kJ/mol and - 117.26 kJ/mol with Cdk5/p25. Overall, we recommend two natural product candidate inhibitors to target the pharmacological inhibition of Cdk5/p25 in tau-associated neurological disorders.

Identification of novel acetylcholinesterase inhibitors designed by pharmacophore-based virtual screening, molecular docking and bioassay

In this study, pharmacophore based 3D QSAR models for human acetylcholinesterase (AChE) inhibitors were generated, with good significance, statistical values (r2training = 0.73) and predictability (q2training = 0.67). It was further validated by three methods (Fischer's test, decoy set and Güner-Henry scoring method) to show that the models can be used to predict the biological activities of compounds without costly and time-consuming synthesis. The criteria for virtual screening were also validated by testing the selective AChE inhibitors. Virtual screening experiments and subsequent in vitro evaluation of promising hits revealed a novel and selective AChE inhibitor. Thus, the findings reported herein may provide a new strategy for the discovery of selective AChE inhibitors. The IC50 value of compounds 5c and 6a presented selective inhibition of AChE without inhibiting butyrylcholinesterase (BChE) at uM level. Molecular docking studies were performed to explain the potent AChE inhibition of the target compounds studies to explain high affinity.

Discovery of CNS-Like D3R-Selective Antagonists Using 3D Pharmacophore Guided Virtual Screening

The dopamine D3 receptor is an important CNS target for the treatment of a variety of neurological diseases. Selective dopamine D3 receptor antagonists modulate the improvement of psychostimulant addiction and relapse. In this study, five and six featured pharmacophore models of D3R antagonists were generated and evaluated with the post-hoc score combining two survival scores of active and inactive. Among the Top 10 models, APRRR215 and AHPRRR104 were chosen based on the coefficient of determination (APRRR215: R²training = 0.80; AHPRRR104: R²training = 0.82) and predictability (APRRR215: Q²test = 0.73, R²predictive = 0.82; AHPRRR104: Q²test = 0.86, R²predictive = 0.74) of their 3D-quantitative structure⁻activity relationship models. Pharmacophore-based virtual screening of a large compound library from eMolecules (>3 million compounds) using two optimal models expedited the search process by a 100-fold speed increase compared to the docking-based screening (HTVS scoring function in Glide) and identified a series of hit compounds having promising novel scaffolds. After the screening, docking scores, as an adjuvant predictor, were added to two fitness scores (from the pharmacophore models) and predicted Ki (from PLSs of the QSAR models) to improve accuracy. Final selection of the most promising hit compounds were also evaluated for CNS-like properties as well as expected D3R antagonism.

The comparison of automated clustering algorithms for resampling representative conformer ensembles with RMSD matrix

Background

The accuracy of any 3D-QSAR, Pharmacophore and 3D-similarity based chemometric target fishing models are highly dependent on a reasonable sample of active conformations. Since a number of diverse conformational sampling algorithm exist, which exhaustively generate enough conformers, however model building methods relies on explicit number of common conformers.

Results

In this work, we have attempted to make clustering algorithms, which could find reasonable number of representative conformer ensembles automatically with asymmetric dissimilarity matrix generated from openeye tool kit. RMSD was the important descriptor (variable) of each column of the N × N matrix considered as N variables describing the relationship (network) between the conformer (in a row) and the other N conformers. This approach used to evaluate the performance of the well-known clustering algorithms by comparison in terms of generating representative conformer ensembles and test them over different matrix transformation functions considering the stability. In the network, the representative conformer group could be resampled for four kinds of algorithms with implicit parameters. The directed dissimilarity matrix becomes the only input to the clustering algorithms.

Conclusions

Dunn index, Davies–Bouldin index, Eta-squared values and omega-squared values were used to evaluate the clustering algorithms with respect to the compactness and the explanatory power. The evaluation includes the reduction (abstraction) rate of the data, correlation between the sizes of the population and the samples, the computational complexity and the memory usage as well. Every algorithm could find representative conformers automatically without any user intervention, and they reduced the data to 14–19% of the original values within 1.13 s per sample at the most. The clustering methods are simple and practical as they are fast and do not ask for any explicit parameters. RCDTC presented the maximum Dunn and omega-squared values of the four algorithms in addition to consistent reduction rate between the population size and the sample size. The performance of the clustering algorithms was consistent over different transformation functions. Moreover, the clustering method can also be applied to molecular dynamics sampling simulation results.

Electronic supplementary material

The online version of this article (doi:10.1186/s13321-017-0208-0) contains supplementary material, which is available to authorized users.