MegaSyn: Integrating Generative Molecular Design, Automated Analog Designer, and Synthetic Viability Prediction.

Generative machine learning models have become widely adopted in drug discovery and other fields to produce new molecules and explore molecular space, with the goal of discovering novel compounds with optimized properties. These generative models are frequently combined with transfer learning or scoring of the physicochemical properties to steer generative design, yet often, they are not capable of addressing a wide variety of potential problems, as well as converge into similar molecular space when combined with a scoring function for the desired properties. In addition, these generated compounds may not be synthetically feasible, reducing their capabilities and limiting their usefulness in real-world scenarios. Here, we introduce a suite of automated tools called MegaSyn representing three components: a new hill-climb algorithm, which makes use of SMILES-based recurrent neural network (RNN) generative models, analog generation software, and retrosynthetic analysis coupled with fragment analysis to score molecules for their synthetic feasibility. We show that by deconstructing the targeted molecules and focusing on substructures, combined with an ensemble of generative models, MegaSyn generally performs well for the specific tasks of generating new scaffolds as well as targeted analogs, which are likely synthesizable and druglike. We now describe the development, benchmarking, and testing of this suite of tools and propose how they might be used to optimize molecules or prioritize promising lead compounds using these RNN examples provided by multiple test case examples.

Prospective Assessment of Virtual Screening Heuristics Derived Using a Novel Fusion Score.

High-throughput screening (HTS) is a widespread method in early drug discovery for identifying promising chemical matter that modulates a target or phenotype of interest. Because HTS campaigns involve screening millions of compounds, it is often desirable to initiate screening with a subset of the full collection. Subsequently, virtual screening methods prioritize likely active compounds in the remaining collection in an iterative process. With this approach, orthogonal virtual screening methods are often applied, necessitating the prioritization of hits from different approaches. Here, we introduce a novel method of fusing these prioritizations and benchmark it prospectively on 17 screening campaigns using virtual screening methods in three descriptor spaces. We found that the fusion approach retrieves 15% to 65% more active chemical series than any single machine-learning method and that appropriately weighting contributions of similarity and machine-learning scoring techniques can increase enrichment by 1% to 19%. We also use fusion scoring to evaluate the tradeoff between screening more chemical matter initially in lieu of replicate samples to prevent false-positives and find that the former option leads to the retrieval of more active chemical series. These results represent guidelines that can increase the rate of identification of promising active compounds in future iterative screens.

Informing the Selection of Screening Hit Series with in Silico Absorption, Distribution, Metabolism, Excretion, and Toxicity Profiles.

High-throughput screening (HTS) has enabled millions of compounds to be assessed for biological activity, but challenges remain in the prioritization of hit series. While biological, absorption, distribution, metabolism, excretion, and toxicity (ADMET), purity, and structural data are routinely used to select chemical matter for further follow-up, the scarcity of historical ADMET data for screening hits limits our understanding of early hit compounds. Herein, we describe a process that utilizes a battery of in-house quantitative structure-activity relationship (QSAR) models to generate in silico ADMET profiles for hit series to enable more complete characterizations of HTS chemical matter. These profiles allow teams to quickly assess hit series for desirable ADMET properties or suspected liabilities that may require significant optimization. Accordingly, these in silico data can direct ADMET experimentation and profoundly impact the progression of hit series. Several prospective examples are presented to substantiate the value of this approach.

Novel oxazolidinone calcitonin gene-related peptide (CGRP) receptor antagonists for the acute treatment of migraine.

In our efforts to develop CGRP receptor antagonists as backups to MK-3207, 2, we employed a scaffold hopping approach to identify a series of novel oxazolidinone-based compounds. The development of a structurally diverse, potent (20, cAMP+HS IC50=0.67 nM), and selective compound (hERG IC50=19 µM) with favorable rodent pharmacokinetics (F=100%, t1/2=7h) is described. Key to this development was identification of a 3-substituted spirotetrahydropyran ring that afforded a substantial gain in potency (10 to 35-fold).

Synthesis and Evaluation of 5-Fluoro-2-aryloxazolo[5,4-b]pyridines as ß-Amyloid PET Ligands and Identification of MK-3328.

5-Fluoro-2-aryloxazolo[5,4-b]pyridines were synthesized and investigated as potential (18)F containing ß-amyloid PET ligands. In competition binding assays using human AD brain homogenates, compounds 14b, 16b, and 17b were identified as having favorable potency versus human ß-amyloid plaque and were radiolabeled for further evaluation in in vitro binding and in vivo PET imaging experiments. These studies led to the identification of 17b (MK-3328) as a candidate PET ligand for the clinical assessment of ß-amyloid plaque load.

Discovery of triarylethanolamine inhibitors of the Kv1.5 potassium channel.

A series of triarylethanolamine inhibitors of the Kv1.5 potassium channel have been prepared and evaluated for their effects in vitro and in vivo. The structure-activity relationship (SAR) studies described herein led to the development of potent, selective and orally active inhibitors of Kv1.5.

Analogs of MK-499 are differentially affected by a mutation in the S6 domain of the hERG K+ channel.

Drug-induced long QT syndrome has been principally ascribed to block of the cardiac hERG K(+) channel. Methanesulfonanilides, such as MK-499, E-4031 and dofetilide, are potent hERG antagonists that likely bind along the S6 helix within the inner vestibule of the pore. To further investigate these interactions, we broadly explored the structure-activity relationships of closely related analogs of MK-499 using a high-throughput ion flux assay, and evaluated in greater detail using patch-clamp electrophysiology. We observed that substitutions at the 4-position on the benzopyran ring significantly affected the potency of these analogs with the rank order of unsubstituted approximately ketone>amine>hydroxyl, implicating an important interaction at this position. We also evaluated the potency of these analogs on an S6 mutant of hERG (F656A) previously shown to significantly reduce the affinity for MK-499 and other known hERG antagonists (e.g. cisapride, terfenadine). In contrast to MK-499 (4-hydroxyl) and either the amine or unsubstituted analogs, the potency of the ketone analog was unaffected by this mutation suggesting that a compensatory interaction may be unveiled with the aromatic to apolar substitution, possibly through hydrogen bonding with Ser624 based on molecular modeling. More significantly, we found that this mutation rendered hERG susceptible to block in the closed-state by the smaller, unsubstituted analog, but not by MK-499 or larger analogs. Together these data suggest that interaction with Phe656 is not an absolute requirement for the binding of all methanesulfonanilide compounds, and that this residue may play a broader role in regulating access to the inner vestibule.

Reagent Selector: using Synthon Analysis to visualize reagent properties and assist in combinatorial library design.

Reagent Selector is an intranet-based tool that aids in the selection of reagents for use in combinatorial library construction. The user selects an appropriate reagent group as a query, for example, primary amines, and further refines it on the basis of various physicochemical properties, resulting in a list of potential reagents. The results of this selection process are, in turn, converted into synthons: the fragments or R-groups that are to be incorporated into the combinatorial library. The Synthon Analysis interface graphically depicts the chemical properties for each synthon as a function of the topological bond distance from the scaffold attachment point. Displayed in this fashion, the user is able to visualize the property space for the universe of synthons as well as that of the synthons selected. Ultimately, the reagent list that embodies the selected synthons is made available to the user for reagent procurement. Application of the approach to a sample reagent list for a G-protein coupled receptor targeted library is described.

Web enabling technology for the design, enumeration, optimization and tracking of compound libraries.

Motivated by the need to augment Merck's in-house small molecule collection, web-based tools for designing, enumerating, optimizing and tracking compound libraries have been developed. The path leading to the current version of this Virtual Library Tool Kit (VLTK) is discussed in context of the (then) available commercial offerings and the constraints and requirements imposed by the end users. Though the effort was initiated to simplify the tasks of designing novel, drug-like and diverse compound libraries containing between 2K-10K unique entities, it has also evolved into a powerful tool for outsourcing syntheses as well as lead identification and optimization. The web tool includes components that select reagents, analyze synthons, identify backup reagents, enumerate libraries, calculate properties, optimize libraries and finally track the synthesized compounds through biological assays. In addition to accommodating project specific designs and virtual 3D library scanning, the application includes tools for parallel synthesis, laboratory automation and compound registration.

Biochemical and structural characterization of a novel class of inhibitors of the type 1 insulin-like growth factor and insulin receptor kinases.

The type 1 insulin-like growth factor receptor (IGF-1R) is often overexpressed on tumor cells and is believed to play an important role in anchorage-independent proliferation. Additionally, cell culture studies have indicated that IGF-1R confers increased resistance to apoptosis caused by radiation or chemotherapeutic agents. Thus, inhibitors of the intracellular kinase domain of this receptor may have utility for the clinical treatment of cancer. As part of an effort to develop clinically useful inhibitors of IGF-1R kinase, a novel class of pyrrole-5-carboxaldehyde compounds was investigated. The compounds exhibited selectivity against the closely related insulin receptor kinase intrinsically and in cell-based assays. The inhibitors formed a reversible, covalent adduct at the kinase active site, and treatment of such adducts with sodium borohydride irreversibly inactivated the enzyme. Analysis of a tryptic digest of a covalently modified IGF-1R kinase fragment revealed that the active site Lys1003 had been reductively alkylated with the aldehyde inhibitor. Reductive alkylation of the insulin receptor kinase with one of these inhibitors led to a similarly inactivated enzyme which was examined by X-ray crystallography. The crystal structure confirmed the modification of the active site lysine side chain and revealed details of the key interactions between the inhibitor and enzyme.

Macrocyclic piperazinones as potent dual inhibitors of farnesyltransferase and geranylgeranyltransferase-I.

A series of macrocyclic piperazinone compounds with dual farnesyltransferase/geranylgeranyltransferase-I inhibitory activity was prepared. These compounds were found to be potent inhibitors of protein prenylation in cell culture. A hypothesis for the binding mode of compound 3o in FPTase is proposed.

Molecular model of the alpha(IIb)beta(3) integrin.

A molecular model of the alpha(IIb)beta(3) integrin has been developed utilizing (i). the crystal structure of alpha(v)beta(3), (ii). homology model of the alpha(IIb) subdomain, and (iii). the docking of alpha(IIb)beta(3)/alpha(v)beta(3) dual and selective inhibitors into the putative binding sites of alpha(IIb)beta(3) and alpha(v)beta(3). Since the binding sites of these integrins are located at the interface between the two heads of the individual subunits, only the alpha(IIb)beta(3) head region is modeled. The 3D conformations of two loops in alpha(IIb), whose residues have been implicated in non-peptide ligand binding, could not be determined from homology with alpha(v) alone. Mutagenesis data and the modeling of small ligand binding contributed to the rational design of these loop conformations. The final energy minimized loop conformations exhibit permissible phi/psi angles and contribute to a binding site model of alpha(IIb)beta(3) that is consistent with both the known mutagenesis studies and in-house structure-activity relationships. The charged residues alpha(IIb):E117 and beta(3):R214 are found to dominate the ligand-protein binding interaction. The previously identified "exosite" is also identified as a hydrogen bond, hydrophobic or pi-pi interaction with Y190, similar to the recently proposed binding model of alpha(v)beta(3).

Random forest: a classification and regression tool for compound classification and QSAR modeling.

A new classification and regression tool, Random Forest, is introduced and investigated for predicting a compound's quantitative or categorical biological activity based on a quantitative description of the compound's molecular structure. Random Forest is an ensemble of unpruned classification or regression trees created by using bootstrap samples of the training data and random feature selection in tree induction. Prediction is made by aggregating (majority vote or averaging) the predictions of the ensemble. We built predictive models for six cheminformatics data sets. Our analysis demonstrates that Random Forest is a powerful tool capable of delivering performance that is among the most accurate methods to date. We also present three additional features of Random Forest: built-in performance assessment, a measure of relative importance of descriptors, and a measure of compound similarity that is weighted by the relative importance of descriptors. It is the combination of relatively high prediction accuracy and its collection of desired features that makes Random Forest uniquely suited for modeling in cheminformatics.

3-Aminopyrrolidinone farnesyltransferase inhibitors: design of macrocyclic compounds with improved pharmacokinetics and excellent cell potency.

A series of macrocyclic 3-aminopyrrolidinone farnesyltransferase inhibitors (FTIs) has been synthesized. Compared with previously described linear 3-aminopyrrolidinone FTIs such as compound 1, macrocycles such as 49 combined improved pharmacokinetic properties with a reduced potential for side effects. In dogs, oral bioavailability was good to excellent, and increases in plasma half-life were due to attenuated clearance. It was observed that in vivo clearance correlated with the flexibility of the molecules and this concept proved useful in the design of FTIs that exhibited low clearance, such as FTI 78. X-ray crystal structures of compounds 49 and 66 complexed with farnesyltransferase (FTase)-farnesyl diphosphate (FPP) were determined, and they provide details of the key interactions in such ternary complexes. Optimization of this 3-aminopyrrolidinone series of compounds led to significant increases in potency, providing 83 and 85, the most potent inhibitors of FTase in cells described to date.