Gargoyles: An Open Source Graph-Based Molecular Optimization Method Based on Deep Reinforcement Learning.

Automatic optimization methods for compounds in the vast compound space are important for drug discovery and material design. Several machine learning-based molecular generative models for drug discovery have been proposed, but most of these methods generate compounds from scratch and are not suitable for exploring and optimizing user-defined compounds. In this study, we developed a compound optimization method based on molecular graphs using deep reinforcement learning. This method searches for compounds on a fragment-by-fragment basis and at high density by generating fragments to be added atom by atom. Experimental results confirmed that the quantum electrodynamics (QED), the optimization target set in this study, was enhanced by searching around the starting compound. As a use case, we successfully enhanced the activity of a compound by targeting dopamine receptor D2 (DRD2). This means that the generated compounds are not structurally dissimilar from the starting compounds, as well as increasing their activity, indicating that this method is suitable for optimizing molecules from a given compound. The source code is available at https://github.com/sekijima-lab/GARGOYLES.

Discovery of a Hidden Trypanosoma cruzi Spermidine Synthase Binding Site and Inhibitors through In Silico, In Vitro, and X-ray Crystallography.

In drug discovery research, the selection of promising binding sites and understanding the binding mode of compounds are crucial fundamental studies. The current understanding of the proteins-ligand binding model extends beyond the simple lock and key model to include the induced-fit model, which alters the conformation to match the shape of the ligand, and the pre-existing equilibrium model, selectively binding structures with high binding affinity from a diverse ensemble of proteins. Although methods for detecting target protein binding sites and virtual screening techniques using docking simulation are well-established, with numerous studies reported, they only consider a very limited number of structures in the diverse ensemble of proteins, as these methods are applied to a single structure. Molecular dynamics (MD) simulation is a method for predicting protein dynamics and can detect potential ensembles of protein binding sites and hidden sites unobservable in a single-point structure. In this study, to demonstrate the utility of virtual screening with protein dynamics, MD simulations were performed on Trypanosoma cruzi spermidine synthase to obtain an ensemble of dominant binding sites with a high probability of existence. The structure of the binding site obtained through MD simulation revealed pockets in addition to the active site that was present in the initial structure. Using the obtained binding site structures, virtual screening of 4.8 million compounds by docking simulation, in vitro assays, and X-ray analysis was conducted, successfully identifying two hit compounds.

Toxicokinetic analysis of the anticoagulant rodenticides warfarin & diphacinone in Egyptian fruit bats (Rousettus aegyptiacus) as a comparative sensitivity assessment for Bonin fruit bats (Pteropus pselaphon).

Anticoagulant rodenticides have been widely used to eliminate wild rodents, which as invasive species on remote islands can disturb ecosystems. Since rodenticides can cause wildlife poisoning, it is necessary to evaluate the sensitivity of local mammals and birds to the poisons to ensure the rodenticides are used effectively. The Bonin Islands are an archipelago located 1000 km southeast of the Japanese mainland and are famous for the unique ecosystems. Here the first-generation anticoagulant rodenticide diphacinone has been used against introduced black rats (Rattus rattus). The only land mammal native to the archipelago is the Bonin fruit bat (Pteropus pselaphon), but little is known regarding its sensitivity to rodenticides. In this study, the Egyptian fruit bats (Rousettus aegyptiacus) was used as a model animal for in vivo pharmacokinetics and pharmacodynamics analysis and in vitro enzyme kinetics using their hepatic microsomal fractions. The structure of vitamin K epoxide reductase (VKORC1), the target protein of the rodenticide in the Bonin fruit bat, was predicted from its genome and its binding affinity to rodenticides was evaluated. The Egyptian fruit bats excreted diphacinone slowly and showed similar sensitivity to rats. In contrast, they excreted warfarin, another first-generation rodenticide, faster than rats and recovered from the toxic effect faster. An in silico binding study also indicated that the VKORC1 of fruit bats is relatively tolerant to warfarin, but binds strongly to diphacinone. These results suggest that even chemicals with the same mode of action display different sensitivities in different species: fruit bat species are relatively resistant to warfarin, but vulnerable to diphacinone.

Screening for Inhibitors of Main Protease in SARS-CoV-2: In Silico and In Vitro Approach Avoiding Peptidyl Secondary Amides.

In addition to vaccines, antiviral drugs are essential for suppressing COVID-19. Although several inhibitor candidates were reported for SARS-CoV-2 main protease, most are highly polar peptidomimetics with poor oral bioavailability and cell membrane permeability. Here, we conducted structure-based virtual screening and in vitro assays to obtain hit compounds belonging to a new chemical space, excluding peptidyl secondary amides. In total, 180 compounds were subjected to the primary assay at 20 µM, and nine compounds with inhibition rates of >5% were obtained. The IC50 of six compounds was determined in dose-response experiments, with the values on the order of 10-4 M. Although nitro groups were enriched in the substructure of the hit compounds, they did not significantly contribute to the binding interaction in the predicted docking poses. Physicochemical properties prediction showed good oral absorption. These new scaffolds are promising candidates for future optimization.

Statistical potentials for RNA-protein interactions optimized by CMA-ES.

Characterizing RNA-protein interactions remains an important endeavor, complicated by the difficulty in obtaining the relevant structures. Evaluating model structures via statistical potentials is in principle straight-forward and effective. However, given the relatively small size of the existing learning set of RNA-protein complexes optimization of such potentials continues to be problematic. Notably, interaction-based statistical potentials have problems in addressing large RNA-protein complexes. In this study, we adopted a novel strategy with covariance matrix adaptation (CMA-ES) to calculate statistical potentials, successfully identifying native docking poses.

Effect of charged mutation on aggregation of a pentapeptide: Insights from molecular dynamics simulations.

Aggregation of therapeutic monoclonal antibodies (mAbs) can negatively affect their chemistry, manufacturing, and control attributes and lead to undesirable immune responses in patients. Therefore, optimization of lead mAb drug candidates during discovery stages to mitigate aggregation is increasingly becoming an integral part of their developability assessments. The disruption of short sequence motifs called aggregation prone regions (APRs) found in amino acid sequences of mAb candidates can potentially mitigate their aggregation. In this work, we have performed molecular dynamics simulations to study the aggregation of an APR (VLVIY) found in λ light chains of human antibodies and its single point mutant KLVIY. Eighteen different multicopy peptide simulation systems of "VLVIY" and "KLVIY" were constructed by varying their concentrations, temperatures, termini capping, and flanking gate-keeper regions. Within 20 ns of the simulation, peptide "VLVIY" formed an aggregate of 100 peptides at ~0.1 M concentration with a 60% reduction in solvent accessible surface area (SASA). Furthermore, analysis of the SASA change, peptide cluster distribution, and water residence time demonstrated how Val ➔ Lys mutation resists aggregation and improves solubility. Presence of Lys slows down aggregation kinetics via charge-charge repulsions and by raising the kinetic barrier to formation of large oligomers. However, the effect of the Val ➔ Lys mutation is dependent on sequence and structural contexts around the APR. This mutation also alters the solvation shell around the peptide by favoring solute-solvent interactions, thereby increasing its solubility. This work has provided a detailed mechanistic explanation of how APR disruption can mitigate aggregation in biotherapeutics and improve their developability.

MERMAID: an open source automated hit-to-lead method based on deep reinforcement learning.

The hit-to-lead process makes the physicochemical properties of the hit molecules that show the desired type of activity obtained in the screening assay more drug-like. Deep learning-based molecular generative models are expected to contribute to the hit-to-lead process. The simplified molecular input line entry system (SMILES), which is a string of alphanumeric characters representing the chemical structure of a molecule, is one of the most commonly used representations of molecules, and molecular generative models based on SMILES have achieved significant success. However, in contrast to molecular graphs, during the process of generation, SMILES are not considered as valid SMILES. Further, it is quite difficult to generate molecules starting from a certain molecule, thus making it difficult to apply SMILES to the hit-to-lead process. In this study, we have developed a SMILES-based generative model that can be generated starting from a certain molecule. This method generates partial SMILES and inserts it into the original SMILES using Monte Carlo Tree Search and a Recurrent Neural Network. We validated our method using a molecule dataset obtained from the ZINC database and successfully generated molecules that were both well optimized for the objectives of the quantitative estimate of drug-likeness (QED) and penalized octanol-water partition coefficient (PLogP) optimization. The source code is available at https://github.com/sekijima-lab/mermaid .

Computer aided drug discovery review for infectious diseases with case study of anti-Chagas project.

Neglected tropical diseases (NTDs) are parasitic and bacterial infections that are widespread, especially in the tropics, and cause health problems for about one billion people over 149 countries worldwide. However, in terms of therapeutic agents, for example, nifurtimox and benznidazole were developed in the 1960s to treat Chagas disease, but new drugs are desirable because of their side effects. Drug discovery takes 12 to 14 years and costs $2.6 billon dollars, and hence, computer aided drug discovery (CADD) technology is expected to reduce the time and cost. This paper describes our methods and results based on CADD, mainly for NTDs. An overview of databases, molecular simulation and pharmacophore modeling, contest-based drug discovery, and machine learning and their results are presented herein.

Identification of key interactions between SARS-CoV-2 main protease and inhibitor drug candidates.

The number of cases of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection (COVID-19) has reached over 114,000. SARS-CoV-2 caused a pandemic in Wuhan, China, in December 2019 and is rapidly spreading globally. It has been reported that peptide-like anti-HIV-1 drugs are effective against SARS-CoV Main protease (Mpro). Due to the close phylogenetic relationship between SARS-CoV and SARS-CoV-2, their main proteases share many structural and functional features. Thus, these drugs are also regarded as potential drug candidates targeting SARS-CoV-2 Mpro. However, the mechanism of action of SARS-CoV-2 Mpro at the atomic-level is unknown. In the present study, we revealed key interactions between SARS-CoV-2 Mpro and three drug candidates by performing pharmacophore modeling and 1 µs molecular dynamics (MD) simulations. His41, Gly143, and Glu166 formed interactions with the functional groups that were common among peptide-like inhibitors in all MD simulations. These interactions are important targets for potential drugs against SARS-CoV-2 Mpro.

A prospective compound screening contest identified broader inhibitors for Sirtuin 1.

Potential inhibitors of a target biomolecule, NAD-dependent deacetylase Sirtuin 1, were identified by a contest-based approach, in which participants were asked to propose a prioritized list of 400 compounds from a designated compound library containing 2.5 million compounds using in silico methods and scoring. Our aim was to identify target enzyme inhibitors and to benchmark computer-aided drug discovery methods under the same experimental conditions. Collecting compound lists derived from various methods is advantageous for aggregating compounds with structurally diversified properties compared with the use of a single method. The inhibitory action on Sirtuin 1 of approximately half of the proposed compounds was experimentally accessed. Ultimately, seven structurally diverse compounds were identified.

Molecular Dynamics Simulation reveals the mechanism by which the Influenza Cap-dependent Endonuclease acquires resistance against Baloxavir marboxil.

Baloxavir marboxil (BXM), an antiviral drug for influenza virus, inhibits RNA replication by binding to RNA replication cap-dependent endonuclease (CEN) of influenza A and B viruses. Although this drug was only approved by the FDA in October 2018, drug resistant viruses have already been detected from clinical trials owing to an I38 mutation of CEN. To investigate the reduction of drug sensitivity by the I38 mutant variants, we performed a molecular dynamics (MD) simulation on the CEN-BXM complex structure to analyze variations in the mode of interaction. Our simulation results suggest that the side chain methyl group of I38 in CEN engages in a CH-pi interaction with the aromatic ring of BXM. This interaction is abolished in various I38 mutant variants. Moreover, MD simulation on various mutation models and binding free energy prediction by MM/GBSA method suggest that the I38 mutation precludes any interaction with the aromatic ring of BXA and thereby reduces BXA sensitivity.

Improved Method of Structure-Based Virtual Screening via Interaction-Energy-Based Learning.

Virtual screening is a promising method for obtaining novel hit compounds in drug discovery. It aims to enrich potentially active compounds from a large chemical library for further biological experiments. However, the accuracy of current virtual screening methods is insufficient. In this study, we develop a new virtual screening method named Similarity of Interaction Energy VEctor Score (SIEVE-Score), in which protein-ligand interaction energies are extracted to represent docking poses for machine learning. SIEVE-Score offers substantial improvements compared to other state-of-the-art virtual screening methods, namely, other machine-learning-based scoring functions, interaction fingerprints, and docking software, for the enrichment factor 1% results on the Directory of Useful Decoys, Enhanced (DUD-E). The screening results are also human-interpretable in the form of important interactions for distinguishing between active and inactive compounds. The source code is available at https://github.com/sekijima-lab/SIEVE-Score .

Compound property enhancement by virtual compound synthesis.

During drug discovery, drug candidates are narrowed down over several steps to develop pharmaceutical products. The theoretical chemical space in such steps is estimated to be [Formula: see text]. To cover that space, extensive virtual compound libraries have been developed; however, the compilation of extensive libraries comes at large computational cost. Thus, to reduce the computational cost, researchers have constructed custom-made virtual compound libraries that focus on target diseases. In this study, we develop a system that generates virtual compound libraries from input compounds. When a user inputs a compound, the system recursively applies virtual synthetic reaction rules to the compound to improve its properties. The synthetic pathway can also be traced by the user because the reaction rules in this system are based on real organic synthesis reactions. This system has useful functions for effective drug design, such as structural preservation, allowing the substructures necessary for potency to be maintained. In this paper, to confirm the effect of directional reaction sets, we applied the reaction sets to 100 compounds. Moreover, to confirm that the system can reproduce real synthetic pathways, the synthetic pathways of Ibuprofen and Ofloxacin were explored by inputting isobutyl benzene and 7,8-difluoro-2,3-dihydro-3-methyl-4H-benzoxazine. This application is available at the following URL: http://enh.sekijima-lab.org .

Exploring the selectivity of inhibitor complexes with Bcl-2 and Bcl-XL: A molecular dynamics simulation approach.

B-cell lymphoma 2 (Bcl-2) family proteins are potential drug targets in cancer and have a relatively flat and flexible binding site. ABT-199 is one of the most promising selective Bcl-2 inhibitors, and A-1155463 selectively inhibits Bcl-XL. Although the amino acid sequences of the binding sites of these two inhibitors are similar, the inhibitors selectively bind the target protein. In order to determine the origin of the selectivity of these inhibitors, we conducted molecular dynamics simulations using protein-inhibitor modeling. We confirmed that ASP103 of Bcl-2 is a key residue and that hydrogen bonding between ASP103 and ABT-199 confers the Bcl-2 selectivity of this inhibitor. For Bcl-XL selectivity, the secondary structure of α-helix 3 is a key factor. PHE105, SER106, and LEU108 in the loose α-helix 3 interact with A-1155463 to confer Bcl-XL selectivity. These findings provide important insights into the molecular mechanisms of selective inhibitors of Bcl-2 family proteins.

An iterative compound screening contest method for identifying target protein inhibitors using the tyrosine-protein kinase Yes.

We propose a new iterative screening contest method to identify target protein inhibitors. After conducting a compound screening contest in 2014, we report results acquired from a contest held in 2015 in this study. Our aims were to identify target enzyme inhibitors and to benchmark a variety of computer-aided drug discovery methods under identical experimental conditions. In both contests, we employed the tyrosine-protein kinase Yes as an example target protein. Participating groups virtually screened possible inhibitors from a library containing 2.4 million compounds. Compounds were ranked based on functional scores obtained using their respective methods, and the top 181 compounds from each group were selected. Our results from the 2015 contest show an improved hit rate when compared to results from the 2014 contest. In addition, we have successfully identified a statistically-warranted method for identifying target inhibitors. Quantitative analysis of the most successful method gave additional insights into important characteristics of the method used.

In silico, in vitro, X-ray crystallography, and integrated strategies for discovering spermidine synthase inhibitors for Chagas disease.

Chagas disease results from infection by Trypanosoma cruzi and is a neglected tropical disease (NTD). Although some treatment drugs are available, their use is associated with severe problems, including adverse effects and limited effectiveness during the chronic disease phase. To develop a novel anti-Chagas drug, we virtually screened 4.8 million small molecules against spermidine synthase (SpdSyn) as the target protein using our super computer "TSUBAME2.5" and conducted in vitro enzyme assays to determine the half-maximal inhibitory concentration values. We identified four hit compounds that inhibit T. cruzi SpdSyn (TcSpdSyn) by in silico and in vitro screening. We also determined the TcSpdSyn-hit compound complex structure using X-ray crystallography, which shows that the hit compound binds to the putrescine-binding site and interacts with Asp171 through a salt bridge.

Identification of potential inhibitors based on compound proposal contest: Tyrosine-protein kinase Yes as a target.

A search of broader range of chemical space is important for drug discovery. Different methods of computer-aided drug discovery (CADD) are known to propose compounds in different chemical spaces as hit molecules for the same target protein. This study aimed at using multiple CADD methods through open innovation to achieve a level of hit molecule diversity that is not achievable with any particular single method. We held a compound proposal contest, in which multiple research groups participated and predicted inhibitors of tyrosine-protein kinase Yes. This showed whether collective knowledge based on individual approaches helped to obtain hit compounds from a broad range of chemical space and whether the contest-based approach was effective.

Pharmacophore modeling for anti-Chagas drug design using the fragment molecular orbital method.

BACKGROUND: Chagas disease, caused by the parasite Trypanosoma cruzi, is a neglected tropical disease that causes severe human health problems. To develop a new chemotherapeutic agent for the treatment of Chagas disease, we predicted a pharmacophore model for T. cruzi dihydroorotate dehydrogenase (TcDHODH) by fragment molecular orbital (FMO) calculation for orotate, oxonate, and 43 orotate derivatives. METHODOLOGY/PRINCIPAL FINDINGS: Intermolecular interactions in the complexes of TcDHODH with orotate, oxonate, and 43 orotate derivatives were analyzed by FMO calculation at the MP2/6-31G level. The results indicated that the orotate moiety, which is the base fragment of these compounds, interacts with the Lys43, Asn67, and Asn194 residues of TcDHODH and the cofactor flavin mononucleotide (FMN), whereas functional groups introduced at the orotate 5-position strongly interact with the Lys214 residue. CONCLUSIONS/SIGNIFICANCE: FMO-based interaction energy analyses revealed a pharmacophore model for TcDHODH inhibitor. Hydrogen bond acceptor pharmacophores correspond to Lys43 and Lys214, hydrogen bond donor and acceptor pharmacophores correspond to Asn67 and Asn194, and the aromatic ring pharmacophore corresponds to FMN, which shows important characteristics of compounds that inhibit TcDHODH. In addition, the Lys214 residue is not conserved between TcDHODH and human DHODH. Our analysis suggests that these orotate derivatives should preferentially bind to TcDHODH, increasing their selectivity. Our results obtained by pharmacophore modeling provides insight into the structural requirements for the design of TcDHODH inhibitors and their development as new anti-Chagas drugs.