Predictions of Colloidal Molecular Aggregation Using AI/ML Models.

To facilitate the triage of hits from small molecule screens, we have used various AI/ML techniques and experimentally observed data sets to build models aimed at predicting colloidal aggregation of small organic molecules in aqueous solution. We have found that Naïve Bayesian and deep neural networks outperform logistic regression, recursive partitioning tree, support vector machine, and random forest techniques by having the lowest balanced error rate (BER) for the test set. Derived predictive classification models consistently and successfully discriminated aggregator molecules from nonaggregator hits. An analysis of molecular descriptors in favor of colloidal aggregation confirms previous observations (hydrophobicity, molecular weight, and solubility) in addition to undescribed molecular descriptors such as the fraction of sp3 carbon atoms (Fsp3), and electrotopological state of hydroxyl groups (ES_Sum_sOH). Naïve Bayesian modeling and scaffold tree analysis have revealed chemical features/scaffolds contributing the most to colloidal aggregation and nonaggregation, respectively. These results highlight the importance of scaffolds with high Fsp3 values in promoting nonaggregation. Matched molecular pair analysis (MMPA) has also deciphered context-dependent substitutions, which can be used to design nonaggregator molecules. We found that most matched molecular pairs have a neutral effect on aggregation propensity. We have prospectively applied our predictive models to assist in chemical library triage for optimal plate selection diversity and purchase for high throughput screening (HTS) in drug discovery projects.

Application of artificial intelligence and machine learning techniques to the analysis of dynamic protein sequences.

We apply methods of Artificial Intelligence and Machine Learning to protein dynamic bioinformatics. We rewrite the sequences of a large protein data set, containing both folded and intrinsically disordered molecules, using a representation developed previously, which encodes the intrinsic dynamic properties of the naturally occurring amino acids. We Fourier analyze the resulting sequences. It is demonstrated that classification models built using several different supervised learning methods are able to successfully distinguish folded from intrinsically disordered proteins from sequence alone. It is further shown that the most important sequence property for this discrimination is the sequence mobility, which is the sequence averaged value of the residue-specific average alpha carbon B factor. This is in agreement with previous work, in which we have demonstrated the central role played by the sequence mobility in protein dynamic bioinformatics and biophysics. This finding opens a path to the application of dynamic bioinformatics, in combination with machine learning algorithms, to a range of significant biomedical problems.

Identification and pharmacological characterization of 3,6-diazabicyclo[3.1.1]heptane-3-carboxamides as novel ligands for the α4ß2 and α6/α3ß2ß3 nicotinic acetylcholine receptors (nAChRs).

We have synthesized a novel series of compounds, 3,6-diazabicyclo[3.1.1]heptane-3-carboxamides, targeting both the α4ß2 and α6/α3ß2ß3 nAChRs. Members of the obtained chemical library are partial or full agonists at both the high sensitivity (α4)2(ß2)3 and α6/α3ß2ß3 nAChRs. 3-(Cyclopropylcarbonyl)-3,6-diazabicyclo[3.1.1]heptane (TC-8831 or compound 7 herein) demonstrated a safe in vitro pharmacological profile and the potential for reducing or preventing L-dopa-induced dyskinesias (LID) in several in vivo animal models [1-4]. In vivo metabolism studies in rat and in vitro metabolism studies in liver microsomes from human, rat, dog and monkey showed TC-8831 to be relatively stable. In vivo pharmacokinetic analysis in the rat confirmed brain penetration, with an average brain:plasma ratio of approximately 0.3 across time points from 0.5 to 4 h. Docking into homology models predicted alternative binding modes for TC-8831 and highlighted the importance of the cationic center, hydrogen-bond acceptor, and hydrophobic aliphatic features in promoting binding affinity to both nAChRs. Pharmacophore elucidation confirmed the importance of these key interactions. QSAR modeling suggested that binding affinity is primarily driven by ligand shape, relative positive charge distribution onto the molecular surface, and molecular flexibility. Of the two subtypes, ligand binding to α6ß2ß3 appears to be more sensitive to bulkiness and flexibility.

Comparative study on the use of docking and Bayesian categorization to predict ligand binding to nicotinic acetylcholine receptors (nAChRs) subtypes.

We have carried out a comparative study between docking into homology models and Bayesian categorization, as applied to virtual screening of nicotinic ligands for binding at various nAChRs subtypes (human and rat α4ß2, α7, α3ß4, and α6ß2ß3). We found that although results vary with receptor subtype, Bayesian categorization exhibits higher accuracy and enrichment than unconstrained docking into homology models. However, docking accuracy is improved when one sets up a hydrogen-bond (HB) constraint between the cationic center of the ligand and the main-chain carbonyl group of the conserved Trp-149 or its homologue (a residue involved in cation-π interactions with the ligand basic nitrogen atom). This finding suggests that this HB is a hallmark of nicotinic ligands binding to nAChRs. Best predictions using either docking or Bayesian were obtained with the human α7 nAChR, when 100 nM was used as cutoff for biological activity. We also found that ligand-based Bayesian-derived enrichment factors and structure-based docking-derived enrichment factors highly correlate to each other. Moreover, they correlate with the mean molecular fractional polar surface area of actives ligands and the fractional hydrophobic/hydrophilic surface area of the binding site, respectively. This result is in agreement with the fact that hydrophobicity strongly contributes in promoting nicotinic ligands binding to their cognate nAChRs.

Computational studies of novel carbonyl-containing diazabicyclic ligands interacting with α4ß2 nicotinic acetylcholine receptor (nAChR) reveal alternative binding modes.

We have carried out computational studies on interactions of diazabicyclic amide analogs with α4ß2 nAChR using homology modeling, docking and pharmacophore elucidation techniques. We have found alternative ligand binding modes in most cases. All these diverse poses exhibit the quintessential hydrogen-bonding interaction between the ligand basic nitrogen and the backbone carbonyl oxygen atom of the highly conserved Trp-149. This hydrogen bond was always found to be shorter than the one contracted by the ligand carbonyl group and a second hydrogen-bond made by the cationic center with Tyr-93 of the principal face of the protein. In most of the poses observed, cation-π interactions involved three aromatic residues located in the principal face of the protein: Trp-149, Tyr-190 and Tyr-197. The latter amino acid residue appears to often donate a hydrogen-bond to the ligand carbonyl oxygen atom. We also describe two rings of alternative receptor-based hydrogen-bond donor features equidistantly separated from the carbonyl oxygen of the highly conserved Trp-149 approximately by 5 and 8Å, respectively. These findings could be exploited to design diverse and selective novel chemical libraries for the treatment of diseases and conditions where the α4ß2 nAChR is disrupted, such as Alzheimer disease, Parkinson's disease and l-dopa-induced dyskinesia (LID).

QM-polarized ligand docking accurately predicts the trend in binding affinity of a series of arylmethylene quinuclidine-like derivatives at the α4ß2 and α3ß4 nicotinic acetylcholine receptors (nAChRs).

Compounds containing a quinuclidine scaffold are promising drug candidates for pharmacological management of the central nervous system (CNS) pathologies implicating nAChRs. We have carried out binding affinity and in-silico docking studies of arylmethylene quinuclidine-like derivatives at the α4ß2 receptor using in-vitro receptor binding assay and comparative modeling, respectively. We found that introducing a hydrogen-bond acceptor into the 3-benzylidene quinuclidine derivative resulted in a 266-fold increase in binding affinity and confers agonism properties. By contrast, addition of a phenyl group to 3-benzylidene quinuclidine derivative only results in an 18-fold increase in binding affinity, without conferring agonism. We also found that docking into the orthosteric binding site of the α4ß2 nAChR is consistent with the fact that the basic nitrogen atom donates a hydrogen-bond to the carbonyl group of the highly conserved Trp-149, as initially observed by Dougherty and co-workers.(1) The experimentally-observed trend in binding affinity at both α4ß2 and α3ß4 nAChRs was accurately and independently confirmed by quantum mechanics (QM)-polarized docking. The reduction in binding affinity to the α3ß4 subtype primarily results from a dampening of both coulombic and cation-π interactions.

Novel nicotinic acetylcholine receptor agonists containing carbonyl moiety as a hydrogen bond acceptor.

A novel series of α4ß2 nAChR agonists lacking common pyridine or its bioisosteric heterocycle have been disclosed. Essential pharmacophoric elements of the series are exocyclic carbonyl moiety as a hydrogen bond acceptor and secondary amino group within diaza- or azabicyclic scaffold. Computer modeling studies suggested that molecular shape of the ligand also contributes to promotion of agonism. Proof of concept for improving working memory performance in a novel object recognition task has been demonstrated on a representative of the series, 3-propionyl-3,7-diazabicyclo[3.3.0]octane (34).

Pharmacological properties and predicted binding mode of arylmethylene quinuclidine-like derivatives at the α3ß4 nicotinic acetylcholine receptor (nAChR).

We have carried out a pharmacological evaluation of arylmethylene quinuclidine derivatives interactions with human α3ß4 nAChRs subtype, using cell-based receptor binding, calcium-influx, electrophysiological patch-clamp assays and molecular modeling techniques. We have found that the compounds bind competitively to the α3ß4 receptor with micromolar affinities and some of the compounds behave as non-competitive antagonists (compounds 1, 2 and 3), displaying submicromolar IC(50) values. These evidences suggest a mixed mode of action for these compounds, having interactions at the orthosteric site and more pronounced interactions at an allosteric site to block agonist effects. One of the compounds, 1-benzyl-3-(diphenylmethylene)-1-azoniabicyclo[2.2.2]octane chloride (compound 3), exhibited poorly reversible use-dependent block of α3ß4 channels. We also found that removal of a phenyl group from compound 1 confers a partial agonism to the derived analog (compound 6). Introducing a hydrogen-bond acceptor into the 3-benzylidene quinuclidine derivative (compound 7) increases agonism potency at the α3ß4 receptor subtype. Docking into the orthosteric binding site of a α3ß4 protein structure derived by comparative modeling accurately predicted the experimentally-observed trend in binding affinity. Results supported the notion that binding requires a hydrogen bond formation between the ligand basic nitrogen and the backbone carbonyl oxygen atom of the conserved Trp-149.

3D molecular descriptors important for clinical success.

The pharmacokinetic and safety profiles of clinical drug candidates are greatly influenced by their requisite physicochemical properties. In particular, it has been shown that 2D molecular descriptors such as fraction of Sp3 carbon atoms (Fsp3) and number of stereo centers correlate with clinical success. Using the proteomic off-target hit rate of nicotinic ligands, we found that shape-based 3D descriptors such as the radius of gyration and shadow indices discriminate off-target promiscuity better than do Fsp3 and the number of stereo centers. We have deduced the relevant descriptor values required for a ligand to be nonpromiscuous. Investigating the MDL Drug Data Report (MDDR) database as compounds move from the preclinical stage toward the market, we have found that these shape-based 3D descriptors predict clinical success of compounds at preclinical and phase1 stages vs compounds withdrawn from the market better than do Fsp3 and LogD. Further, these computed 3D molecular descriptors correlate well with experimentally observed solubility, which is among well-known physicochemical properties that drive clinical success. We also found that about 84% of launched drugs satisfy either Shadow index or Fsp3 criteria, whereas withdrawn and discontinued compounds fail to meet the same criteria. Our studies suggest that spherical compounds (rather than their elongated counterparts) with a minimal number of aromatic rings may exhibit a high propensity to advance from clinical trials to market.

Discovery of (2S,3R)-N-[2-(pyridin-3-ylmethyl)-1-azabicyclo[2.2.2]oct-3-yl]benzo[b]furan-2-carboxamide (TC-5619), a selective α7 nicotinic acetylcholine receptor agonist, for the treatment of cognitive disorders.

(2S,3R)-N-[2-(Pyridin-3-ylmethyl)-1-azabicyclo[2.2.2]oct-3-yl]benzo[b]furan-2-carboxamide (7a, TC-5619), a novel selective agonist of the α7 neuronal nicotinic acetylcholine receptor, has been identified as a promising drug candidate for the treatment of cognitive impairment associated with neurological disorders. 7a demonstrated more than a thousand-fold separation between the affinities for the α7 and α4ß2 receptor subtypes and had no detectable effects on muscle or ganglionic nicotinic receptor subtypes, indicating a marked selectivity for the central nervous system over the peripheral nervous system. Results obtained from homology modeling and docking explain the observed selectivity. 7a had positive effects across cognitive, positive, and negative symptoms of schizophrenia in animal models and was additive or synergistic with the antipsychotic clozapine. Compound 7a, as an augmentation therapy to the standard treatment with antipsychotics, demonstrated encouraging results on measures of negative symptoms and cognitive dysfunction in schizophrenia and was well tolerated in a phase II clinical proof of concept trial in patients with schizophrenia.

Comparison of acetylcholine receptor interactions of the marine toxins, 13-desmethylspirolide C and gymnodimine.

The interaction of 13-desmethylspirolide C (SPX-desMe-C) and gymnodimine with several nicotinic and muscarinic acetylcholine receptors was investigated. Interaction at the muscarinic receptors was minimal. At nicotinic receptors, both SPX-desMe-C and gymnodimine displayed greatest affinity for the α7 receptor. The rank order for binding affinity (Ki) for SPX-desMe-C was α7 > α6ß3ß4α5 >> rat α3ß4, α1ßγδ > α4ß4, human α3ß4 > human α4ß2 > rat α4ß2 and for gymnodimine was α7, α6ß3ß4α5 > rat α3ß4 > human α3ß4, α4ß4 > rat α4ß2, human α4ß2 > α1ßγδ. Both molecules antagonized agonist-induced nicotinic responses. The antagonism rank order of potency (IC(50)) for SPX-desMe-C was α7 > low sensitivity (LS) α4ß2 > human α3ß4 > high sensitivity (HS) α4ß2, α1ßγδ > α4ß4 > rat α3ß4 and for gymnodimine was LS α4ß2 > human α3ß4 > α7 > HS α4ß2 > α4ß4 > rat α3ß4 > α1ßγδ. Neither gymnodimine nor SPX-desMe-C antagonism could be surmounted by increasing concentrations of nicotine. To elucidate the nature of this insurmountable blockade, we carried out homology modelling and molecular docking studies of both ligands with α7 nAChR. Their very high binding affinity results from very tight hydrophobic enclosures, in addition to previously reported hydrogen-bond and cation-π interactions. Also, the higher the hydrophilic surface area of the binding site of nAChRs, the weaker the binding affinity of both ligands. Together these results show the targets of action are nicotinic and define these marine toxins as additional tools to advance our understanding regarding interactions between antagonists and the nAChR ligand binding domain.

Discovery of novel α7 nicotinic acetylcholine receptor ligands via pharmacophoric and docking studies of benzylidene anabaseine analogs.

Based on pharmacophore elucidation and docking studies on interactions of benzylidene anabaseine analogs with AChBPs and α7 nAChR, novel spirodiazepine and spiroimidazoline quinuclidine series have been designed. Binding studies revealed that some of hydrogen-bond donor containing compounds exhibit improved affinity and selectivity for the α7 nAChR subtype in comparison with most potent metabolite of GTS-21, 3-(4-hydroxy-2-methoxybenzylidene)-anabaseine. Hydrophobicity and rigidity of the ligand also contribute into its binding affinity. We also describe alternative pharmacophoric features equidistant from the carbonyl oxygen atom of the conserved Trp-148 of the principal face, which may be exploited to further design diverse focused libraries targeting the α7 nAChR.

Docking studies of benzylidene anabaseine interactions with α7 nicotinic acetylcholine receptor (nAChR) and acetylcholine binding proteins (AChBPs): application to the design of related α7 selective ligands.

AChBPs isolated from Lymnaea stagnalis (Ls), Aplysia californica (Ac) and Bulinus truncatus (Bt) have been extensively used as structural prototypes to understand the molecular mechanisms that underlie ligand-interactions with nAChRs [1]. Here, we describe docking studies on interactions of benzylidene anabaseine analogs with AChBPs and α7 nAChR. Results reveal that docking of these compounds using Glide software accurately reproduces experimentally-observed binding modes of DMXBA and of its active metabolite, in the binding pocket of Ac. In addition to the well-known nicotinic pharmacophore (positive charge, hydrogen-bond acceptor, and hydrophobic aromatic groups), a hydrogen-bond donor feature contributes to binding of these compounds to Ac, Bt, and the α7 nAChR. This is consistent with benzylidene anabaseine analogs with OH and NH(2) functional groups showing the highest binding affinity of these congeners, and the position of the ligand shown in previous X-ray crystallographic studies of ligand-Ac complexes. In the predicted ligand-Ls complex, by contrast, the ligand OH group acts as hydrogen-bond acceptor. We have applied our structural findings to optimizing the design of novel spirodiazepine and spiroimidazoline quinuclidine series. Binding and functional studies revealed that these hydrogen-bond donor containing compounds exhibit improved affinity and selectivity for the α7 nAChR subtype and demonstrate partial agonism. The gain in affinity is also due to conformational restriction, tighter hydrophobic enclosures, and stronger cation-π interactions. The use of AChBPs structure as a surrogate to predict binding affinity to α7 nAChR has also been investigated. On the whole, we found that molecular docking into Ls binding site generally scores better than when a α7 homology model, Bt or Ac crystal structure is used.