Predicting the Likelihood of Molecules to Act as Modulators of Protein-Protein Interactions.

Targeting protein-protein interactions (PPIs) by small molecule modulators (iPPIs) is an attractive strategy for drug therapy, and some iPPIs have already been introduced into the clinic. Blocking PPIs is however considered to be a more difficult task than inhibiting enzymes or antagonizing receptor activity. In this paper, we examine whether it is possible to predict the likelihood of molecules to act as iPPIs. Using our in-house iterative stochastic elimination (ISE) algorithm, we constructed two classification models that successfully distinguish between iPPIs from the iPPI-DB database and decoy molecules from either the Enamine HTS collection (ISE 1) or the ZINC database (ISE 2). External test sets of iPPIs taken from the TIMBAL database and decoys from Enamine HTS or ZINC were screened by the models: the area under the curve for the receiver operating characteristic curve was 0.85-0.89, and the Enrichment Factor increased from an initial 1 to as much as 66 for ISE 1 and 57 for ISE 2. Screening of the Enamine HTS and ZINC data sets through both models results in a library of ∼1.3 million molecules that pass either one of the models. This library is enriched with iPPI candidates that are structurally different from known iPPIs, and thus, it is useful for target-specific screenings and should accelerate the discovery of iPPI drug candidates. The entire library is available in Table S6.

Activity Models of Key GPCR Families in the Central Nervous System: A Tool for Many Purposes.

G protein-coupled receptors (GPCRs) are targets of many drugs, of which ∼25% are indicated for central nervous system (CNS) disorders. Drug promiscuity affects their efficacy and safety profiles. Predicting the polypharmacology profile of compounds against GPCRs can thus provide a basis for producing more precise therapeutics by considering the targets and the anti-targets in that family of closely related proteins. We provide a tool for predicting the polypharmacology of compounds within prominent GPCR families in the CNS: serotonin, dopamine, histamine, muscarinic, opioid, and cannabinoid receptors. Our in-house algorithm, "iterative stochastic elimination" (ISE), produces high-quality ligand-based models for agonism and antagonism at 31 GPCRs. The ISE models correctly predict 68% of CNS drug-GPCR interactions, while the "similarity ensemble approach" predicts only 33%. The activity models correctly predict 56% of reported activities of DrugBank molecules for these CNS receptors. We conclude that the combination of interactions and activity profiles generated by screening through our models form the basis for subsequent designing and discovering novel therapeutics, either single, multitargeting, or repurposed.

Dual Inhibitors of AChE and BACE-1 for Reducing Aß in Alzheimer's Disease: From In Silico to In Vivo.

Alzheimer's disease (AD) is a complex and widespread condition, still not fully understood and with no cure yet. Amyloid beta (Aß) peptide is suspected to be a major cause of AD, and therefore, simultaneously blocking its formation and aggregation by inhibition of the enzymes BACE-1 (ß-secretase) and AChE (acetylcholinesterase) by a single inhibitor may be an effective therapeutic approach, as compared to blocking one of these targets or by combining two drugs, one for each of these targets. We used our ISE algorithm to model each of the AChE peripheral site inhibitors and BACE-1 inhibitors, on the basis of published data, and constructed classification models for each. Subsequently, we screened large molecular databases with both models. Top scored molecules were docked into AChE and BACE-1 crystal structures, and 36 Molecules with the best weighted scores (based on ISE indexes and docking results) were sent for inhibition studies on the two enzymes. Two of them inhibited both AChE (IC50 between 4-7 µM) and BACE-1 (IC50 between 50-65 µM). Two additional molecules inhibited only AChE, and another two molecules inhibited only BACE-1. Preliminary testing of inhibition by F681-0222 (molecule 2) on APPswe/PS1dE9 transgenic mice shows a reduction in brain tissue of soluble Aß42.

Computational design of substrate selective inhibition.

Most enzymes act on more than a single substrate. There is frequently a need to block the production of a single pathogenic outcome of enzymatic activity on a substrate but to avoid blocking others of its catalytic actions. Full blocking might cause severe side effects because some products of that catalysis may be vital. Substrate selectivity is required but not possible to achieve by blocking the catalytic residues of an enzyme. That is the basis of the need for "Substrate Selective Inhibitors" (SSI), and there are several molecules characterized as SSI. However, none have yet been designed or discovered by computational methods. We demonstrate a computational approach to the discovery of Substrate Selective Inhibitors for one enzyme, Prolyl Oligopeptidase (POP) (E.C 3.4.21.26), a serine protease which cleaves small peptides between Pro and other amino acids. Among those are Thyrotropin Releasing Hormone (TRH) and Angiotensin-III (Ang-III), differing in both their binding (Km) and in turnover (kcat). We used our in-house "Iterative Stochastic Elimination" (ISE) algorithm and the structure-based "Pharmacophore" approach to construct two models for identifying SSI of POP. A dataset of ~1.8 million commercially available molecules was initially reduced to less than 12,000 which were screened by these models to a final set of 20 molecules which were sent for experimental validation (five random molecules were tested for comparison). Two molecules out of these 20, one with a high score in the ISE model, the other successful in the pharmacophore model, were confirmed by in vitro measurements. One is a competitive inhibitor of Ang-III (increases its Km), but non-competitive towards TRH (decreases its Vmax).

Prediction and Experimental Confirmation of Novel Peripheral Cannabinoid-1 Receptor Antagonists.

Small molecules targeting peripheral CB1 receptors have therapeutic potential in a variety of disorders including obesity-related, hormonal, and metabolic abnormalities, while avoiding the psychoactive effects in the central nervous system. We applied our in-house algorithm, iterative stochastic elimination, to produce a ligand-based model that distinguishes between CB1R antagonists and random molecules by physicochemical properties only. We screened ∼2 million commercially available molecules and found that about 500 of them are potential candidates to antagonize the CB1R. We applied a few criteria for peripheral activity and narrowed that set down to 30 molecules, out of which 15 could be purchased. Ten out of those 15 showed good affinity to the CB1R and two of them with nanomolar affinities (Ki of ∼400 nM). The eight molecules with top affinities were tested for activity: two compounds were pure antagonists, and five others were inverse agonists. These molecules are now being examined in vivo for their peripheral versus central distribution and subsequently will be tested for their effects on obesity in small animals.

Structural basis of glycogen branching enzyme deficiency and pharmacologic rescue by rational peptide design.

Glycogen branching enzyme 1 (GBE1) plays an essential role in glycogen biosynthesis by generating α-1,6-glucosidic branches from α-1,4-linked glucose chains, to increase solubility of the glycogen polymer. Mutations in the GBE1 gene lead to the heterogeneous early-onset glycogen storage disorder type IV (GSDIV) or the late-onset adult polyglucosan body disease (APBD). To better understand this essential enzyme, we crystallized human GBE1 in the apo form, and in complex with a tetra- or hepta-saccharide. The GBE1 structure reveals a conserved amylase core that houses the active centre for the branching reaction and harbours almost all GSDIV and APBD mutations. A non-catalytic binding cleft, proximal to the site of the common APBD mutation p.Y329S, was found to bind the tetra- and hepta-saccharides and may represent a higher-affinity site employed to anchor the complex glycogen substrate for the branching reaction. Expression of recombinant GBE1-p.Y329S resulted in drastically reduced protein yield and solubility compared with wild type, suggesting this disease allele causes protein misfolding and may be amenable to small molecule stabilization. To explore this, we generated a structural model of GBE1-p.Y329S and designed peptides ab initio to stabilize the mutation. As proof-of-principle, we evaluated treatment of one tetra-peptide, Leu-Thr-Lys-Glu, in APBD patient cells. We demonstrate intracellular transport of this peptide, its binding and stabilization of GBE1-p.Y329S, and 2-fold increased mutant enzymatic activity compared with untreated patient cells. Together, our data provide the rationale and starting point for the screening of small molecule chaperones, which could become novel therapies for this disease.

Discovering Novel and Diverse Iron-Chelators in Silico.

Specific iron chelation is a validated strategy in anticancer drug discovery. However, only a few chemical classes (4-5 categories) have been reported to date. We discovered in silico five new structurally diverse iron-chelators by screening through models based on previously known chelators. To encompass a larger chemical space and propose newer scaffolds, we used our iterative stochastic elimination (ISE) algorithm for model building and subsequent virtual screening (VS). The ISE models were developed by training a data set of 130 reported iron-chelators. The developed models are statistically significant with area under the receiver operating curve greater than 0.9. The models were used to screen the Enamine chemical database of ∼1.8 million molecules. The top ranked 650 molecules were reduced to 50 diverse structures, and a few others were eliminated due to the presence of reactive groups. Finally, 34 molecules were purchased and tested in vitro. Five compounds were identified with significant iron-chelation activity in Cal-G assay. Intracellular iron-chelation study revealed one compound as equivalent in potency to the iron chelating "gold standards" deferoxamine and deferiprone. The amount of discovered positives (5 out of 34) is expected by the realistic enrichment factor of the model.

Computational Discovery and Experimental Confirmation of TLR9 Receptor Antagonist Leads.

Toll-like receptors (TLR) are receptors of innate immunity that recognize pathogen associated molecular patterns. They play a critical role in many pathological states, in acute and chronic inflammatory processes. TLR9 is a promising target for drug discovery, since it has been implicated in several pathologies, including defense against viral infections and psoriasis. Immune-modulators are promising molecules for therapeutic intervention in these indications. TLR9 is located in the endosome and activated by dsDNA with CpG motives encountered in microbial DNA. Here we report on a combined approach to discover new TLR9 antagonists by computational chemistry and cell based assays. We used our in-house iterative stochastic elimination (ISE) algorithm to create models that distinguish between TLR9 antagonists ("actives") and other molecules ("inactives"), based on molecular physicochemical properties. Subsequent screening and scoring of a data set of 1.8 million commercially available molecules led to the purchasing of top scored molecules, which were tested in a new cell based system based on human pattern recognition receptors (PRRs) stably expressed in NIH3T3 fibroblasts. As described previously, this cell line shows a very low endogenous PRR-activity and contains a reporter gene which is selectively activated by the integrated human PRR enabling rapid screening of potential ligands. IC50 values of each of these top scored molecules were determined. Out of 60 molecules tested, 56 showed antagonistic activity. We discovered 21 new highly potential antagonists with IC50 values lower than 10 µM, with 5 of them having IC50 values under 1 µM.

Candidate Therapeutics by Screening for Multitargeting Ligands: Combining the CB2 Receptor With CB1, PPARγ and 5-HT4 Receptors.

In recent years, the cannabinoid type 2 receptor (CB2R) has become a major target for treating many disease conditions. The old therapeutic paradigm of "one disease-one target-one drug" is being transformed to "complex disease-many targets-one drug." Multitargeting, therefore, attracts much attention as a promising approach. We thus focus on designing single multitargeting agents (MTAs), which have many advantages over combined therapies. Using our ligand-based approach, the "Iterative Stochastic Elimination" (ISE) algorithm, we produce activity models of agonists and antagonists for desired therapeutic targets and anti-targets. These models are used for sequential virtual screening and scoring large libraries of molecules in order to pick top-scored candidates for testing in vitro and in vivo. In this study, we built activity models for CB2R and other targets for combinations that could be used for several indications. Those additional targets are the cannabinoid 1 receptor (CB1R), peroxisome proliferator-activated receptor gamma (PPARγ), and 5-Hydroxytryptamine receptor 4 (5-HT4R). All these models have high statistical parameters and are reliable. Many more CB2R/CBIR agonists were found than combined CB2R agonists with CB1R antagonist activity (by 200 fold). CB2R agonism combined with PPARγ or 5-HT4R agonist activity may be used for treating Inflammatory Bowel Disease (IBD). Combining CB2R agonism with 5-HT4R generates more candidates (14,008) than combining CB2R agonism with agonists for the nuclear receptor PPARγ (374 candidates) from an initial set of ∼2.1 million molecules. Improved enrichment of true vs. false positives may be achieved by requiring a better ISE score cutoff or by performing docking. Those candidates can be purchased and tested experimentally to validate their activity. Further, we performed docking to CB2R structures and found lower statistical performance of the docking ("structure-based") compared to ISE modeling ("ligand-based"). Therefore, ISE modeling may be a better starting point for molecular discovery than docking.

In silico Docking Analysis for Blocking JUNO-IZUMO1 Interaction Identifies Two Small Molecules that Block in vitro Fertilization.

Combined hormone drugs are the basis for orally administered contraception. However, they are associated with severe side effects that are even more impactful for women in developing countries, where resources are limited. The risk of side effects may be reduced by non-hormonal small molecules which specifically target proteins involved in fertilization. In this study, we present a virtual docking experiment directed to discover molecules that target the crucial fertilization interactions of JUNO (oocyte) and IZUMO1 (sperm). We docked 913,000 molecules to two crystal structures of JUNO and ranked them on the basis of energy-related criteria. Of the 32 tested candidates, two molecules (i.e., Z786028994 and Z1290281203) demonstrated fertilization inhibitory effect in both an in vitro fertilization (IVF) assay in mice and an in vitro penetration of human sperm into hamster oocytes. Despite this clear effect on fertilization, these two molecules did not show JUNO-IZUMO1 interaction blocking activity as assessed by AVidity-based EXtracellular Interaction Screening (AVEXIS). Therefore, further research is required to determine the mechanism of action of these two fertilization inhibitors.

Human-Based Immune Responsive In Vitro Infection Models for Validation of Novel TLR4 Antagonists Identified by Computational Discovery.

Infectious diseases are still a major problem worldwide. This includes microbial infections, with a constant increase in resistance to the current anti-infectives employed. Toll-like receptors (TLRs) perform a fundamental role in pathogen recognition and activation of the innate immune response. Promising new approaches to combat infections and inflammatory diseases involve modulation of the host immune system via TLR4. TLR4 and its co-receptors MD2 and CD14 are required for immune response to fungal and bacterial infection by recognition of microbial cell wall components, making it a prime target for drug development. To evaluate the efficacy of anti-infective compounds early on, we have developed a series of human-based immune responsive infection models, including immune responsive 3D-skin infection models for modeling fungal infections. By using computational methods: pharmacophore modeling and molecular docking, we identified a set of 46 potential modulators of TLR4, which were screened in several tests systems of increasing complexity, including immune responsive 3D-skin infection models. We could show a strong suppression of cytokine and chemokine response induced by lipopolysacharide (LPS) and Candida albicans for individual compounds. The development of human-based immune responsive assays provides a more accurate and reliable basis for development of new anti-inflammatory or immune-modulating drugs.

Molecular properties from conformational ensembles. 1. Dipole moments of molecules with multiple internal rotations.

We present a novel method for constructing the stable conformational space of small molecules with many rotatable bonds that uses our iterative stochastic elimination (ISE) algorithm, a robust stochastic search method capable of finding ensembles of best solutions for large combinatorial problems. To validate the method, we show that ISE reproduces the best conformers found in a fully exhaustive search, as well as compare computed dipole moments to experimental values, based on molecular ensembles and their Boltzmann distributions. Results were also compared to the alternative molecular dynamics and simulated annealing methods. Our results clarify that many low energy conformations may be required to reproduce molecular properties, while single low energy conformers or ensembles of low energy conformers cannot account for the experimental properties of flexible molecules. Whereas ISE well reproduces conformations that are not separated by very large energy barriers, it has not been successful in reproducing conformations of strained molecules.

Flexible protein-protein docking based on Best-First search algorithm.

We developed a new high resolution protein-protein docking method based on Best-First search algorithm that loosely imitates protein-protein associations. The method operates in two stages: first, we perform a rigid search on the unbound proteins. Second, we search alternately on rigid and flexible degrees of freedom starting from multiple configurations from the rigid search. Both stages use heuristics added to the energy function, which causes the proteins to rapidly approach each other and remain adjacent, while optimizing on the energy. The method deals with backbone flexibility explicitly by searching over ensembles of conformations generated before docking. We ran the rigid docking stage on 66 complexes and grouped the results into four classes according to evaluation criteria used in Critical Assessment of Predicted Interactions (CAPRI; "high," "medium," "acceptable," and "incorrect"). Our method found medium binding conformations for 26% of the complexes and acceptable for additional 44% among the top 10 configurations. Considering all the configurations, we found medium binding conformations for 55% of the complexes and acceptable for additional 39% of the complexes. Introducing side-chains flexibility in the second stage improves the best found binding conformation but harms the ranking. However, introducing side-chains and backbone flexibility improve both the best found binding conformation and the best found conformation in the top 10. Our approach is a basis for incorporating multiple flexible motions into protein-protein docking and is of interest even with the current use of a simple energy function.

Predicting oral druglikeness by iterative stochastic elimination.

Integration of computational methods in the early stages of drug discovery has been one of the key trends in the pharmaceutical industry. Starting with high quality drug candidates should ultimately minimize clinical attrition rates and give rise to higher success rates. In this paper, we present a novel approach for indexing oral druglikeness of compounds. With the Iterative Stochastic Elimination (ISE) Algorithm, we distinguish between orally available drugs and nondrugs by generating sets of optimized descriptors' ranges, each set constituting a "filter". We delineate in this paper how to produce an ensemble of best k-descriptor sets out of the huge number of possibilities, and how to construct a "filter bank" that retains diverse filters by clustering. Finally, we define the "orally bioavailable drug-like" character of individual molecules by combining the filters into an "Orally Bioavailable Druglike Index" (OB-DLI) which may be used to prioritize molecules in databases and discuss its uses in several potential applications. The predictive power with sets of 4-6 descriptors is high (i.e., one filter of 5 descriptors retrieved 81% true positives and >77% true negatives). Thus, OB-DLI has advantages over binary decisions (that use only one filter) not only in raising discriminative power but also in ranking drug candidates according to their chance to be successful oral drugs. We demonstrate the ability of our approach to discover molecular entities with the required property, orally bioavailable drug likeness, that are structurally dissimilar to those of the training set. Comparison of this ISE application to some of the current main methods for classification reveals that our approach has >13% improvement in the Matthews Correlation Coefficient, which measures the success of identifying true and false positives and negatives.

Discovering highly selective and diverse PPAR-delta agonists by ligand based machine learning and structural modeling.

PPAR-δ agonists are known to enhance fatty acid metabolism, preserving glucose and physical endurance and are suggested as candidates for treating metabolic diseases. None have reached the clinic yet. Our Machine Learning algorithm called "Iterative Stochastic Elimination" (ISE) was applied to construct a ligand-based multi-filter ranking model to distinguish between confirmed PPAR-δ agonists and random molecules. Virtual screening of 1.56 million molecules by this model picked ~2500 top ranking molecules. Subsequent docking to PPAR-δ structures was mainly evaluated by geometric analysis of the docking poses rather than by energy criteria, leading to a set of 306 molecules that were sent for testing in vitro. Out of those, 13 molecules were found as potential PPAR-δ agonist leads with EC50 between 4-19 nM and 14 others with EC50 below 10 µM. Most of the nanomolar agonists were found to be highly selective for PPAR-δ and are structurally different than agonists used for model building.

High quality binding modes in docking ligands to proteins.

Multiple near-optimal conformations of protein-ligand complexes provide a better chance for accurate representation of biomolecular interactions, compared with a single structure. We present ISE-dock--a docking program which is based on the iterative stochastic elimination (ISE) algorithm. ISE eliminates values that consistently lead to the worst results, thus optimizing the search for docking poses. It constructs large sets of such poses with no additional computational cost compared with single poses. ISE-dock is validated using 81 protein-ligand complexes from the PDB and its performance was compared with those of Glide, GOLD, and AutoDock. ISE-dock has a better chance than the other three to find more than 60% top single poses under RMSD = 2.0 A and more than 80% under RMSD = 3.0 A from experimental. ISE alone produced at least one 3.0 A or better solutions among the top 20 poses in the entire test set. In 98% of the examined molecules, ISE produced solutions that are closer than 2.0 A from experimental. Paired t-tests (PTT) were used throughout to assess the significance of comparisons between the performances of the different programs. ISE-dock provides more than 100-fold docking solutions in a similar time frame as LGA in AutoDock. We demonstrate the usefulness of the large near optimal populations of ligand poses by showing a correlation between the docking results and experiments that support multiple binding modes in p38 MAP kinase (Pargellis et al., Nat Struct Biol 2002;9:268-272] and in Human Transthyretin (Hamilton, Benson, Cell Mol Life Sci 2001;58:1491-1521).

Constructing ensembles of flexible fragments in native proteins by iterative stochastic elimination is relevant to protein-protein interfaces.

We investigate the extent to which ensembles of flexible fragments (FF), generated by our loop conformational search method, include conformations that are near experimental and reflect conformational changes that these FFs undergo when binary protein-protein complexes are formed. Twenty-eight FFs, which are located in protein-protein interfaces and have different conformations in the bound structure (BS) and unbound structure (UbS) were extracted. The conformational space of these fragments in the BS and UbS was explored with our method which is based on the iterative stochastic elimination (ISE) algorithm. Conformational search of BSs generated bound ensembles and conformational search of UbSs produced unbound ensembles. ISE samples conformations near experimental (less than 1.05 A root mean square deviation, RMSD) for 51 out of the 56 examined fragments in the bound and unbound ensembles. In 14 out of the 28 unbound fragments, it also samples conformations within 1.05 A from the BS in the unbound ensemble. Sampling the bound conformation in the unbound ensemble demonstrates the potential biological relevance of the predicted ensemble. The 10 lowest energy conformations are the best choice for docking experiments, compared with any other 10 conformations of the ensembles. We conclude that generating conformational ensembles for FFs with ISE is relevant to FF conformations in the UbS and BS. Forming ensembles of the isolated proteins with our method prior to docking represents more comprehensively their inherent flexibility and is expected to improve docking experiments compared with results obtained by docking only UbSs.

New drug candidates for liposomal delivery identified by computer modeling of liposomes' remote loading and leakage.

Remote drug loading into nano-liposomes is in most cases the best method for achieving high concentrations of active pharmaceutical ingredients (API) per nano-liposome that enable therapeutically viable API-loaded nano-liposomes, referred to as nano-drugs. This approach also enables controlled drug release. Recently, we constructed computational models to identify APIs that can achieve the desired high concentrations in nano-liposomes by remote loading. While those previous models included a broad spectrum of experimental conditions and dealt only with loading, here we reduced the scope to the molecular characteristics alone. We model and predict API suitability for nano-liposomal delivery by fixing the main experimental conditions: liposome lipid composition and size to be similar to those of Doxil® liposomes. On that basis, we add a prediction of drug leakage from the nano-liposomes during storage. The latter is critical for having pharmaceutically viable nano-drugs. The "load and leak" models were used to screen two large molecular databases in search of candidate APIs for delivery by nano-liposomes. The distribution of positive instances in both loading and leakage models was similar in the two databases screened. The screening process identified 667 molecules that were positives by both loading and leakage models (i.e., both high-loading and stable). Among them, 318 molecules received a high score in both properties and of these, 67 are FDA-approved drugs. This group of molecules, having diverse pharmacological activities, may be the basis for future liposomal drug development.

Inhibition of LDL oxidation by flavonoids in relation to their structure and calculated enthalpy.

Twenty flavonoid compounds of five different subclasses were selected, and the relationship of their structure to the inhibition of low-density lipoprotein (LDL) oxidation in vitro was investigated. The most effective inhibitors, by either copper ion or 2,2'-azobis (2-amidino-propane) dihydrochloride (AAPH) induction, were flavonols and/or flavonoids with two adjacent hydroxyl groups at ring B. In the presence of the later catechol group, the contribution of the double bond and the carbonyl group at ring C was negligible. Isoflavonoids were more effective inhibitors than other flavonoid subclasses with similar structure. Substituting ring B with hydroxyl group(s) at 2' position resulted in a significantly higher inhibitory effect than by substituting ring A or ring B at other positions. The type of LDL inducer had no effect in flavonoids with catechol structure. Calculated heat of formation data (deltadeltaH(f)) revealed that the donation of a hydrogen atom from position 3 was the most likely result, followed by that of a hydroxyl from ring B. Position 3 was favored only in the presence of conjugated double bonds between ring A to ring B. This study makes it possible to assign the contribution of different functional groups among the flavonoid subclasses to in vitro inhibition of LDL oxidation.