ADMET Evaluation in Drug Discovery. 13. Development of in Silico Prediction Models for P-Glycoprotein Substrates

Employing Automated Machine Learning (AutoML) Methods to Facilitate the In Silico ADMET Properties Prediction

The rationale for using ADMET prediction tools in the early drug discovery paradigm is to guide the design of new compounds with favorable ADMET properties and ultimately minimize the attrition rates of drug failures. Artificial intelligence (AI) in in silico ADMET modeling has gained momentum due to its high-throughput and low-cost attributes. In this study, we developed a machine learning model capable of predicting 11 ADMET properties of chemical compounds. Each model was constructed by combining one of 40 classification algorithms including random forest (RF), extreme gradient boosting (XGB), support vector machine (SVM), and gradient boosting (GB) with one of three predefined hyperparameter configurations. This process can be efficiently performed using automated machine learning (AutoML) methods, which automatically search for the best combination of model algorithms and optimized hyperparameters. We developed optimal predictive models for 11 different ADMET properties using the Hyperopt-sklearn AutoML method. All of the developed models depicted an area under the ROC curve (AUC) >0.8. Furthermore, our developed models outperformed most of the ADMET properties and showed comparable performance in other properties when evaluated on external data sets and compared with published predictive models. Our results support the applicability of AutoML in ADMET prediction and will be helpful for ADMET prediction in early-stage drug discovery.

Norditerpene natural products from subterranean fungi with anti-parasitic activity

Cryptosporidium is a common, waterborne gastrointestinal parasite that causes diarrheal disease worldwide. Currently there are no effective therapeutics to treat cryptosporidiosis in at-risk populations. Since natural products are a known source of anti-parasitic compounds, we screened a library of extracts and pure natural product compounds isolated from bacteria and fungi collected from subterranean environments for activity against Cryptosporidium parvum. Eight structurally related norditerpene lactones isolated from the fungus Oidiodendron truncatum collected from the Soudan Iron mine in Minnesota showed potent activity and were further tested to identify the most active compounds. The availability of a diverse suite of natural structural analogs with varying activities allowed us to determine some structure activity relationships for both anti-parasitic activity as well as cytotoxicity. The two most potent compounds, oidiolactones A and B, had EC50s against intracellular Cryptosporidium parvum of 530 and 240 nM respectively without cytotoxicity to confluent HCT-8 host cells. Both compounds also inhibited the related parasite Toxoplasma gondii. Oidiolactone A was active against asexual, but not sexual, stages of C. parvum, and killed 80% of the parasites within 8 hours of treatment. This compound reduced C. parvum infection by 70% in IFNγ-/- mice, with no signs of toxicity. The high potency, low cytotoxicity, and in vivo activity combined with high production, easy isolation from fungi, and synthetic accessibility make oidiolactones A and B attractive scaffolds for the development of new anti-Cryptosporidium therapeutics.

Efflux and uptake transport and gut microbial reactivation of raloxifene glucuronides

Raloxifene has low bioavailability due to extensive glucuronidation in the intestine and the liver, and its pharmacokinetics is associated with high intra- and interindividual variability. Some of this variability could be explained by the enterohepatic recycling of raloxifene, which is driven by transporter-mediated uptake and efflux and gut microbial deglucuronidation of raloxifene glucuronides. These individual processes involved in raloxifene disposition, however, have not been characterized in full detail. In this study, we evaluated the interactions of raloxifene and its three glucuronide metabolites (raloxifene 4'-glucuronide, raloxifene 6-glucuronide and raloxifene 4',6-diglucuronide) with drug transporters using Sf9 membrane vesicles and HEK293 cells. Additionally, we measured the deglucuronidation of raloxifene glucuronides in human faecal extracts. All raloxifene glucuronides were transported by MRP2 and MRP3, whereas raloxifene monoglucuronides were identified as substrates of OATP1B1, OATP1B3 and OATP2B1. All three raloxifene glucuronides were readily deglucuronidated in the presence of faecal extracts, although with high between-subject variability. The results of this study provide further understanding of the disposition of raloxifene, which can help understand the sources behind the interindividual variability in raloxifene pharmacokinetics.

Comprehensive benchmarking of computational tools for predicting toxicokinetic and physicochemical properties of chemicals

Ensuring the safety of chemicals for environmental and human health involves assessing physicochemical (PC) and toxicokinetic (TK) properties, which are crucial for absorption, distribution, metabolism, excretion, and toxicity (ADMET). Computational methods play a vital role in predicting these properties, given the current trends in reducing experimental approaches, especially those that involve animal experimentation. In the present manuscript, twelve software tools implementing Quantitative Structure-Activity Relationship (QSAR) models were selected for the prediction of 17 relevant PC and TK properties. A total of 41 validation datasets were collected from the literature, curated and used for assessing the models' external predictivity, emphasizing the performance of the models inside the applicability domain. Overall, the results confirmed the adequate predictive performance of the majority of the selected tools, with models for PC properties (R2 average = 0.717) generally outperforming those for TK properties (R2 average = 0.639 for regression, average balanced accuracy = 0.780 for classification). Notably, several of the tools evaluated exhibited good predictivity across different properties and were identified as recurring optimal choices. Moreover, a systematic analysis of the chemical space covered by the external validation datasets confirmed the validity of the collected results for relevant chemical categories (e.g., drugs and industrial chemicals), further increasing the confidence in the overall evaluation. The best performing models were ultimately suggested for each investigated property and proposed as robust computational tools for high-throughput assessment of highly relevant chemical properties. SCIENTIFIC CONTRIBUTION: The present manuscript provides an overview of the state-of-the-art available computational tools for predicting the PC and TK properties of chemicals. The results here offer valuable guidance to researchers, regulatory authorities, and the industry in identifying robust computational tools suitable for predicting relevant chemical properties in the context of chemical design, toxicity and environmental fate assessment.

In Silico Screening of 1,3,4-Thiadiazole Derivatives as Inhibitors of Vascular Endothelial Growth Factor Receptor-2 (VEGFR-2)

Angiogenesis plays a pivotal role in the growth, survival, and metastasis of solid tumors, with Vascular Endothelial Growth Factor Receptor-2 (VEGFR-2) being overexpressed in many human solid tumors, making it an appealing target for anti-cancer therapies. This study aimed to identify potential lead compounds with azole moiety exhibiting VEGFR-2 inhibitory effects. A ligand-based pharmacophore model was constructed using the X-ray crystallographic structure of VEGFR-2 complexed with tivozanib (PDB ID: 4ASE) to screen the ZINC15 database. Following virtual screening, six compounds demonstrated promising docking scores and drug-likeness comparable to tivozanib. These hits underwent detailed pharmacokinetic analysis to assess their absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties. Furthermore, Density Functional Theory (DFT) analysis was employed to investigate the molecular orbital properties of the top hits from molecular docking. Molecular dynamics (MD) simulations were conducted to evaluate the conformational stability of the complexes over a 100 ns run. Results indicated that the compounds (ZINC8914312, ZINC8739578, ZINC8927502, and ZINC17138581) exhibited the most promising lead requirements for inhibiting VEGFR-2 and suppressing angiogenesis in cancer therapy. This integrated approach, combining pharmacophore modeling, molecular docking, ADMET studies, DFT analysis, and MD simulations, provides valuable insights into the identification of potential anti-cancer agents targeting VEGFR-2.

Survey of Pharmaceutical Industry's Best Practices around In Vitro Transporter Assessment and Implications for Drug Development: Considerations from the International Consortium for Innovation and Quality for Pharmaceutical Development Transporter Working Group

The International Consortium for Innovation and Quality in Pharmaceutical Development Transporter Working Group had a rare opportunity to analyze a crosspharma collation of in vitro data and assay methods for the evaluation of drug transporter substrate and inhibitor potential. Experiments were generally performed in accordance with regulatory guidelines. Discrepancies, such as not considering the impact of preincubation for inhibition and free or measured in vitro drug concentrations, may be due to the retrospective nature of the dataset and analysis. Lipophilicity was a frequent indicator of crosstransport inhibition (P-gp, BCRP, OATP1B, and OCT1), with high molecular weight (MW ≥500 Da) also common for OATP1B and BCRP inhibitors. A high level of overlap in in vitro inhibition across transporters was identified for BCRP, OATP1B1, and MATE1, suggesting that prediction of DDIs for these transporters will be common. In contrast, inhibition of OAT1 did not coincide with inhibition of any other transporter. Neutrals, bases, and compounds with intermediate-high lipophilicity tended to be P-gp and/or BCRP substrates, whereas compounds with MW <500 Da tended to be OAT3 substrates. Interestingly, the majority of in vitro inhibitors were not reported to be followed up with a clinical study by the submitting company, whereas those compounds identified as substrates generally were. Approaches to metabolite testing were generally found to be similar to parent testing, with metabolites generally being equally or less potent than parent compounds. However, examples where metabolites inhibited transporters in vitro were identified, supporting the regulatory requirement for in vitro testing of metabolites to enable integrated clinical DDI risk assessment. SIGNIFICANCE STATEMENT: A diverse dataset showed that transporter inhibition often correlated with lipophilicity and molecular weight (>500 Da). Overlapping transporter inhibition was identified, particularly that inhibition of BCRP, OATP1B1, and MATE1 was frequent if the compound inhibited other transporters. In contrast, inhibition of OAT1 did not correlate with the other drug transporters tested.

The Goldilocks paradigm: comparing classical machine learning, large language models, and few-shot learning for drug discovery applications

Recent advances in machine learning (ML) have led to newer model architectures including transformers (large language models, LLMs) showing state of the art results in text generation and image analysis as well as few-shot learning (FSLC) models which offer predictive power with extremely small datasets. These new architectures may offer promise, yet the 'no-free lunch' theorem suggests that no single model algorithm can outperform at all possible tasks. Here, we explore the capabilities of classical (SVR), FSLC, and transformer models (MolBART) over a range of dataset tasks and show a 'goldilocks zone' for each model type, in which dataset size and feature distribution (i.e. dataset "diversity") determines the optimal algorithm strategy. When datasets are small ( < 50 molecules), FSLC tend to outperform both classical ML and transformers. When datasets are small-to-medium sized (50-240 molecules) and diverse, transformers outperform both classical models and few-shot learning. Finally, when datasets are of larger and of sufficient size, classical models then perform the best, suggesting that the optimal model to choose likely depends on the dataset available, its size and diversity. These findings may help to answer the perennial question of which ML algorithm is to be used when faced with a new dataset.

Exploring the dynamics of the ABCB1 membrane transporter P-glycoprotein in the presence of ATP and active/non-active compounds through molecular dynamics simulations

P-glycoprotein (Pgp) is a member of the ATP-binding cassette family of transporters that confers multidrug resistance to cancer cells and is actively involved in the pharmacokinetics and toxicokinetics of a big variety of drugs. Extensive studies have provided insights into the binding of many compounds, but the precise mechanism of translocation across the membrane remains unknown; in this context, the major challenge has been to understand the basis for its polyspecificity. In this study, molecular dynamics (MD) simulations of human P-gp (hP-gp) in an explicit membrane-and-water environment were performed to investigate the dynamic behavior of the transporter in the presence of different compounds (active and inactive) in the binding pocket and ATP molecules within the nucleotide binding domains (NBDs). The complexes studied involve four compounds: cyclosporin A (CSA), amiodarone (AMI), pamidronate (APD), and valproic acid (VPA). While CSA and AMI are known to interact with P-gp, APD and VPA do not. The results highlighted how the presence of ATP notably contributed to increased flexibility of key residues in NBD1 of active systems, indicating potential conformational changes activating the translocation mechanism. MD simulations reveal how these domains adapt and respond to the presence of different substrates, as well as the influence of ATP binding on their flexibility. Furthermore, distinctive behavior was observed in the presence of active and inactive compounds, particularly in the arrangement of ATP between NBDs, supporting the proposed nucleotide sandwich dimer mechanism for ATP binding. This study provides comprehensive insights into P-gp behavior with various ligands and ATP, offering implications for drug development, toxicity assessment and demonstrating the validity of the results derived from the MD simulations.

Potential Effects of Essential Oil from Plinia cauliflora (Mart.) Kausel on Leishmania: In Vivo, In Vitro, and In Silico Approaches

In the search for new chemotherapeutic alternatives for cutaneous leishmaniasis (CL), essential oils are promising due to their diverse biological potential. In this study, we aimed to investigate the chemical composition and leishmanicidal and anti-inflammatory potential of the essential oil isolated from the leaves of Plinia cauliflora (PCEO). The chemical composition of PCEO showed β-cis-Caryophyllene (24.4%), epi-γ-Eudesmol (8%), 2-Naphthalenemethanol[decahydro-alpha] (8%), and trans-Calamenene (6.6%) as its major constituents. Our results showed that the PCEO has moderate cytotoxicity (CC50) of 137.4 and 143.7 μg/mL on mice peritoneal exudate cells (mPEC) and Vero cells, respectively. The PCEO was able to significantly decrease mPEC infection by Leishmania amazonensis and Leishmania braziliensis. The value of the inhibitory concentration (IC50) on amastigote forms was about 7.3 µg/mL (L. amazonensis) and 7.2 µg/mL (L. braziliensis). We showed that PCEO induced drastic ultrastructural changes in both species of Leishmania and had a high selectivity index (SI) > 18. The in silico ADMET analysis pointed out that PCEO can be used for the development of oral and/or topical formulation in the treatment of CL. In addition, we also demonstrated the in vivo anti-inflammatory effect, with a 95% reduction in paw edema and a decrease by at least 21.4% in migration immune cells in animals treated with 50 mg/kg of PCEO. Taken together, our results demonstrate that PCEO is a promising topical therapeutic agent against CL.

Study Models of Drug-Drug Interactions Involving P-Glycoprotein: The Potential Benefit of P-Glycoprotein Modulation at the Kidney and Intestinal Levels

P-glycoprotein (P-gp) is a crucial membrane transporter situated on the cell's apical surface, being responsible for eliminating xenobiotics and endobiotics. P-gp modulators are compounds that can directly or indirectly affect this protein, leading to changes in its expression and function. These modulators can act as inhibitors, inducers, or activators, potentially causing drug-drug interactions (DDIs). This comprehensive review explores diverse models and techniques used to assess drug-induced P-gp modulation. We cover several approaches, including in silico, in vitro, ex vivo, and in vivo methods, with their respective strengths and limitations. Additionally, we explore the therapeutic implications of DDIs involving P-gp, with a special focus on the renal and intestinal elimination of P-gp substrates. This involves enhancing the removal of toxic substances from proximal tubular epithelial cells into the urine or increasing the transport of compounds from enterocytes into the intestinal lumen, thereby facilitating their excretion in the feces. A better understanding of these interactions, and of the distinct techniques applied for their study, will be of utmost importance for optimizing drug therapy, consequently minimizing drug-induced adverse and toxic effects.

Descriptor-Free Deep Learning QSAR Model for the Fraction Unbound in Human Plasma

Chemical-specific parameters are either measured in vitro or estimated using quantitative structure-activity relationship (QSAR) models. The existing body of QSAR work relies on extracting a set of descriptors or fingerprints, subset selection, and training a machine learning model. In this work, we used a state-of-the-art natural language processing model, Bidirectional Encoder Representations from Transformers, which allowed us to circumvent the need for calculation of these chemical descriptors. In this approach, simplified molecular-input line-entry system (SMILES) strings were embedded in a high-dimensional space using a two-stage training approach. The model was first pre-trained on a masked SMILES token task and then fine-tuned on a QSAR prediction task. The pre-training task learned meaningful high-dimensional embeddings based upon the relationships between the chemical tokens in the SMILES strings derived from the "in-stock" portion of the ZINC 15 dataset─a large dataset of commercially available chemicals. The fine-tuning task then perturbed the pre-trained embeddings to facilitate prediction of a specific QSAR endpoint of interest. The power of this model stems from the ability to reuse the pre-trained model for multiple different fine-tuning tasks, reducing the computational burden of developing multiple models for different endpoints. We used our framework to develop a predictive model for fraction unbound in human plasma (fu,p). This approach is flexible, requires minimum domain expertise, and can be generalized for other parameters of interest for rapid and accurate estimation of absorption, distribution, metabolism, excretion, and toxicity.

Ensemble Machine Learning Approaches Based on Molecular Descriptors and Graph Convolutional Networks for Predicting the Efflux Activities of MDR1 and BCRP Transporters

Multidrug resistance (MDR1) and breast cancer resistance protein (BCRP) play important roles in drug absorption and distribution. Computational prediction of substrates for both transporters can help reduce time in drug discovery. This study aimed to predict the efflux activity of MDR1 and BCRP using multiple machine learning approaches with molecular descriptors and graph convolutional networks (GCNs). In vitro efflux activity was determined using MDR1- and BCRP-expressing cells. Predictive performance was assessed using an in-house dataset with a chronological split and an external dataset. CatBoost and support vector regression showed the best predictive performance for MDR1 and BCRP efflux activities, respectively, of the 25 descriptor-based machine learning methods based on the coefficient of determination (R2). The single-task GCN showed a slightly lower performance than descriptor-based prediction in the in-house dataset. In both approaches, the percentage of compounds predicted within twofold of the observed values in the external dataset was lower than that in the in-house dataset. Multi-task GCN did not show any improvements, whereas multimodal GCN increased the predictive performance of BCRP efflux activity compared with single-task GCN. Furthermore, the ensemble approach of descriptor-based machine learning and GCN achieved the highest predictive performance with R2 values of 0.706 and 0.587 in MDR1 and BCRP, respectively, in time-split test sets. This result suggests that two different approaches to represent molecular structures complement each other in terms of molecular characteristics. Our study demonstrated that predictive models using advanced machine learning approaches are beneficial for identifying potential substrate liability of both MDR1 and BCRP.

Pyrrolidinedione-thiazolidinone hybrid molecules with potent cytotoxic effect in squamous cell carcinoma SCC-15 cells

The hybrid heterocyclic molecules are perspective materials in the development of anticancer drugs. Here, the pyrrolidinedione-thiazolidinone hybrid molecules were designed as potent anticancer agents. This study aimed to investigate the cytotoxic effect of three derivatives 1-(4-hydroxyphenyl)-, 1-(4-chlorophenyl)- and 1-(4-bromophenyl)-3-[5-[2-chloro-3-(4-nitrophenyl)prop-2-enylidene]-4-oxo-2-thioxothiazolidine-3-yl]pyrrolidine-2,5-diones (Les-6287, Les-6294, and Les-6328, respectively), their effect on the production of the reactive oxygen species (ROS), apoptosis induction, and expression of genes - PPARγ, AHR, and NRFL2 - whose products are important in metabolism in human tongue squamous cell carcinoma cells of SCC-15 line. The results of resazurin reduction and lactate dehydrogenase (LDH) release assays proved the toxicity of the tested derivatives for the SCC-15 cells. Les-6287, Les-6294, and Les-6328 inhibited the viability of SCC-15 cells with the half-maximal effective concentration (EC50) in the range of 10.18-32.75 µM at 24 and 48 h treatment. These derivatives reduced the metabolism of SCC-15 cells with the half-maximal inhibitory concentration (IC50) of 6.72-39.85 µM at 24 and 48 h treatment. Les-6287, Les-6294, and Les-6328 reduced the metabolism of normal human keratinocytes of HaCaT line murine fibroblasts of Balb/c 3T3 line to a lesser extent. The compounds used in a range from 50 to 100 µM concentrations decreased ROS production in the SCC-15 cells. The derivatives Les-6287 and Les-6328 decreased the level of expression of mRNA of PPARγ, AHR, and NRFL2 genes in these cells at PPARγ siRNA knockdown and without it. Thus, the anticancer effect of studied hybrid pyrrolidinedione-thiazolidinones in the SCC-15 carcinoma cells is accompanied by a reduction of their metabolic activity and ROS level, and increase in caspase 3 activity. However, these changes are not the result of direct interaction of Les-6287, Les-6294, and Les-6328 with the PPARγ molecule.

Identification of in vitro metabolites of an Allium organosulfur compound and environmental toxicity prediction as part of its risk assessment

Propyl-propane-thiosulfonate (PTSO) is an organosulfur compound found inAllium spp. Due to its antioxidant and antimicrobial activities, PTSO has been proposed for applications in the agri-food sector, such as feed additive. However, its use with commercial purposes depends on its toxicity evaluation. The present work aimed to perform a pilot-study of toxicokinetic profile of PTSO combining in silico and in vitro techniques, important steps in the risk assessment process. In silico ecotoxicity studies were also performed considering the importance of the environmental impact of the compound before its commercial use. First, an analytical method has been developed and validated to determine the original compound and its metabolites by ultra-performance liquid chromatography-tandem mass spectrometry. The phase I and II metabolism of PTSO was predicted using Meta-Pred Web Server. For the phase I metabolism, rat (male and female) and human liver microsomes were incubated with PTSO and NADPH regeneration system. Furthermore, in the phase II, microsomes were incubated with PTSO and glutathione or uridine 5'- diphosphoglucuronic acid. The analysis revealed the presence of propylpropane thiosulfinate (PTS) originated by redox reaction in phase I, and two conjugates from the phase II: S-propylmercaptoglutathione (GSSP) and S-propylmercaptocysteine (CSSP). Additionally, considering the environmental fate of PTSO and its metabolites, the ADME parameters and the potential ecotoxicity were also predicted using in silico softwares. The results of the ecotoxicity in silico study evidenced that the metabolism induced the formation of detoxified metabolites from the parent compound, except for dimercaprol and 3-mercaptopropane1,2-diol. Further in vivo assays are needed to confirm this prediction.

Explaining Blood-Brain Barrier Permeability of Small Molecules by Integrated Analysis of Different Transport Mechanisms

The blood-brain barrier (BBB) represents a major obstacle to delivering drugs to the central nervous system (CNS), resulting in the lack of effective treatment for many CNS diseases including brain cancer. To accelerate CNS drug development, computational prediction models could save the time and effort needed for experimental evaluation. Here, we studied BBB permeability focusing on active transport (influx and efflux) as well as passive diffusion using previously published and self-curated data sets. We created prediction models based on physicochemical properties, molecular substructures, or their combination to understand which mechanisms contribute to BBB permeability. Our results show that features that predicted passive diffusion over membranes overlap with features that explain endothelial permeation of approved CNS-active drugs. We also identified physical properties and molecular substructures that positively or negatively predicted BBB transport. These findings provide guidance toward identifying BBB-permeable compounds by optimally matching physicochemical and molecular properties to BBB transport mechanisms.

Identification of 1,2,3-triazole-phthalimide derivatives as potential drugs against COVID-19: a virtual screening, docking and molecular dynamic study

In this work we aimed to perform an in silico predictive screening, docking and molecular dynamic study to identify 1,2,3-triazole-phthalimide derivatives as drug candidates against SARS-CoV-2. The in silico prediction of pharmacokinetic and toxicological properties of hundred one 1,2,3-triazole-phtalimide derivatives, obtained from SciFinder® library, were investigated. Compounds that did not show good gastrointestinal absorption, violated the Lipinski's rules, proved to be positive for the AMES test, and showed to be hepatotoxic or immunotoxic in our ADMET analysis, were filtered out of our study. The hit compounds were further subjected to molecular docking on SARS-CoV-2 target proteins. The ADMET analysis revealed that 43 derivatives violated the Lipinski's rules and 51 other compounds showed to be positive for the toxicity test. Seven 1,2,3-triazole-phthalimide derivatives (A7, A8, B05, E35, E38, E39, and E40) were selected for molecular docking and MFCC-ab initio analysis. The results of molecular docking pointed the derivative E40 as a promising compound interacting with multiple target proteins of SARS-CoV-2. The complex E40-Mpro was found to have minimum binding energy of -10.26 kcal/mol and a general energy balance, calculated by the quantum mechanical analysis, of -8.63 eV. MD simulation and MMGBSA calculations confirmed that the derivatives E38 and E40 have high binding energies of -63.47 ± 3 and -63.31 ± 7 kcal/mol against SARS-CoV-2 main protease. In addition, the derivative E40 exhibited excellent interaction values and inhibitory potential against SAR-Cov-2 main protease and viral nucleocapsid proteins, suggesting this derivative as a potent antiviral for the treatment and/or prophylaxis of COVID-19.Communicated by Ramaswamy H. Sarma.

Interpretable-ADMET: a web service for ADMET prediction and optimization based on deep neural representation

MOTIVATION

In the process of discovery and optimization of lead compounds, it is difficult for non-expert pharmacologists to intuitively determine the contribution of substructure to a particular property of a molecule.

RESULTS

In this work, we develop a user-friendly web service, named interpretable-absorption, distribution, metabolism, excretion and toxicity (ADMET), which predict 59 ADMET-associated properties using 90 qualitative classification models and 28 quantitative regression models based on graph convolutional neural network and graph attention network algorithms. In interpretable-ADMET, there are 250 729 entries associated with 59 kinds of ADMET-associated properties for 80 167 chemical compounds. In addition to making predictions, interpretable-ADMET provides interpretation models based on gradient-weighted class activation map for identifying the substructure, which is important to the particular property. Interpretable-ADMET also provides an optimize module to automatically generate a set of novel virtual candidates based on matched molecular pair rules. We believe that interpretable-ADMET could serve as a useful tool for lead optimization in drug discovery.

AVAILABILITY AND IMPLEMENTATION

Interpretable-ADMET is available at http://cadd.pharmacy.nankai.edu.cn/interpretableadmet/.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Synthesis, molecular docking, and biological evaluation of novel 1,2,4-triazole-isatin derivatives as potential Mycobacterium tuberculosis shikimate kinase inhibitors

The issue of emerging resistance to antitubercular drugs has created a formidable barrier in the effective prevention and cure of tuberculosis globally. In an effort to search for new antimycobacterial agents, possibly comprising new pharmacophore, novel triazole-isatin derivatives were designed as Mycobacterium tuberculosis shikimate kinase inhibitors and synthesized by microwave-assisted method. The synthesized molecules were evaluated for their antimycobacterial activity by MABA assay against M. tuberculosis H37Rv. The molecule 5h demonstrated MIC of 0.8 μg/ml and good safety profile with higher selectivity index with HEK293 cell line. The antimycobacterial activity was further substantiated with molecular docking studies. The triazole-isatin derivatives showed significant binding interactions with amino acid residues in the active site of the enzyme. These studies revealed that molecule 5h could act as a potential lead molecule for further studies to find new target-directed molecules.

A feature transferring workflow between data-poor compounds in various tasks

Compound screening by in silico approaches has advantages in identifying high-activity leading compounds and can predict the safety of the drug. A key challenge is that the number of observations of drug activity and toxicity accumulation varies by target in different datasets, some of which are more understudied than others. Owing to an overall insufficiency and imbalance of drug data, it is hard to accurately predict drug activity and toxicity of multiple tasks by the existing models. To solve this problem, this paper proposed a two-stage transfer learning workflow to develop a novel prediction model, which can accurately predict drug activity and toxicity of the targets with insufficient observations. We built a balanced dataset based on the Tox21 dataset and developed a drug activity and toxicity prediction model based on Siamese networks and graph convolution to produce multitasking output. We also took advantage of transfer learning from data-rich targets to data-poor targets. We showed greater accuracy in predicting the activity and toxicity of compounds to targets with rich data and poor data. In Tox21, a relatively rich dataset, the prediction model accuracy for classification tasks was 0.877 AUROC. In the other five unbalanced datasets, we also found that transfer learning strategies brought the accuracy of models to a higher level in understudied targets. Our models can overcome the imbalance in target data and predict the compound activity and toxicity of understudied targets to help prioritize upcoming biological experiments.

The Application of in silico Methods for Prediction of Blood-Brain Barrier Permeability of Small Molecule PET Tracers

Designing positron emission tomography (PET) tracers for targets in the central nervous system (CNS) is challenging. Besides showing high affinity and high selectivity for their intended target, these tracers have to be able to cross the blood-brain barrier (BBB). Since only a small fraction of small molecules is estimated to be able to cross the BBB, tools that can predict permeability at an early stage during the development are of great importance. One such tool is in silico models for predicting BBB-permeability. Thus far, such models have been built based on CNS drugs, with one exception. Herein, we sought to discuss and analyze if in silico predictions that have been built based on CNS drugs can be applied for CNS PET tracers as well, or if dedicated models are needed for the latter. Depending on what is taken into account in the prediction, i.e., passive diffusion or also active influx/efflux, there may be a need for a model build on CNS PET tracers. Following a brief introduction, an overview of a few selected in silico BBB-permeability predictions is provided along with a short historical background to the topic. In addition, a combination of previously reported CNS PET tracer datasets were assessed in a couple of selected models and guidelines for predicting BBB-permeability. The selected models were either predicting only passive diffusion or also the influence of ADME (absorption, distribution, metabolism and excretion) parameters. To conclude, we discuss the potential need of a prediction model dedicated for CNS PET tracers and present the key issues in respect to setting up a such a model.

The Role of Uptake and Efflux Transporters in the Disposition of Glucuronide and Sulfate Conjugates

Glucuronidation and sulfation are the most typical phase II metabolic reactions of drugs. The resulting glucuronide and sulfate conjugates are generally considered inactive and safe. They may, however, be the most prominent drug-related material in the circulation and excreta of humans. The glucuronide and sulfate metabolites of drugs typically have limited cell membrane permeability and subsequently, their distribution and excretion from the human body requires transport proteins. Uptake transporters, such as organic anion transporters (OATs and OATPs), mediate the uptake of conjugates into the liver and kidney, while efflux transporters, such as multidrug resistance proteins (MRPs) and breast cancer resistance protein (BCRP), mediate expulsion of conjugates into bile, urine and the intestinal lumen. Understanding the active transport of conjugated drug metabolites is important for predicting the fate of a drug in the body and its safety and efficacy. The aim of this review is to compile the understanding of transporter-mediated disposition of phase II conjugates. We review the literature on hepatic, intestinal and renal uptake transporters participating in the transport of glucuronide and sulfate metabolites of drugs, other xenobiotics and endobiotics. In addition, we provide an update on the involvement of efflux transporters in the disposition of glucuronide and sulfate metabolites. Finally, we discuss the interplay between uptake and efflux transport in the intestine, liver and kidneys as well as the role of transporters in glucuronide and sulfate conjugate toxicity, drug interactions, pharmacogenetics and species differences.

Structure-Function Relationships in the Human P-Glycoprotein (ABCB1): Insights from Molecular Dynamics Simulations

P-Glycoprotein (P-gp) is a transmembrane protein belonging to the ATP binding cassette superfamily of transporters, and it is a xenobiotic efflux pump that limits intracellular drug accumulation by pumping compounds out of cells. P-gp contributes to a reduction in toxicity, and has broad substrate specificity. It is involved in the failure of many cancer and antiviral chemotherapies due to the phenomenon of multidrug resistance (MDR), in which the membrane transporter removes chemotherapeutic drugs from target cells. Understanding the details of the ligand–P-gp interaction is therefore critical for the development of drugs that can overcome the MDR phenomenon, for the early identification of P-gp substrates that will help us to obtain a more effective prediction of toxicity, and for the subsequent outdesign of substrate properties if needed. In this work, a series of molecular dynamics (MD) simulations of human P-gp (hP-gp) in an explicit membrane-and-water environment were performed to investigate the effects of binding different compounds on the conformational dynamics of P-gp. The results revealed significant differences in the behaviour of P-gp in the presence of active and non-active compounds within the binding pocket, as different patterns of movement were identified that could be correlated with conformational changes leading to the activation of the translocation mechanism. The predicted ligand–P-gp interactions are in good agreement with the available experimental data, as well as the estimation of the binding-free energies of the studied complexes, demonstrating the validity of the results derived from the MD simulations.

Part-II- in silico drug design: application and success

In silico tools have indeed reframed the steps involved in traditional drug discovery and development process and the term in silico has become a familiar term in pharmaceutical sector like the terms in vitro and in vivo. The successful design of HIV protease inhibitors, Saquinavir, Indinavir and other important medicinal agents, initiated interest of researchers in structure based drug design approaches (SBDD). The interactions between biomolecules and a ligand, binding energy, free energy and stability of biomolecule-ligand complex can be envisioned and predicted by applying molecular docking studies. Protein-ligand, protein-protein, DNA-ligand interactions etc. aid in elucidating molecular level mechanisms of drug molecules. In the Ligand based drug design (LBDD) approaches, QSAR studies have tremendously contributed to the development of antimicrobial, anticancer, antimalarial agents. In the recent years, multiQSAR (mt-QSAR) approaches have been successfully employed for designing drugs against multifactorial diseases. Output of a research in several instances is rewarding when both SBDD and LBDD approaches are combined. Application of in silico studies for prediction of pharmacokinetics was once a real challenge but one can see unlimited number publications comprising tools, data bases which can accurately predict almost all the pharmacokinetic parameters. Absorption, distribution, metabolism, transporters, blood brain barrier permeability, hERG toxicity, P-gp affinity and several toxicological end points can be accurately predicted for a candidate molecule before its synthesis. In silico approaches are greatly encouraged a result of growing limitations and new legislations related to the animal use for research. The combined use of in vitro data and in silico tools will definitely decrease the use of animal testing in the future.In this chapter, in silico approaches and their applications are reviewed and discussed giving suitable examples.

Comparative in silico prediction of P-glycoprotein-mediated transport for 2010-2020 US FDA-approved drugs using six Web-tools

P-glycoprotein (P-gp) is an efflux pump implicated in pharmacokinetics and drug-drug interactions. The identification of its substrates is consequently an important issue, notably for drugs under development. For such a purpose, various in silico methods have been developed, but their relevance remains to be fully established. The present study was designed to get insight about this point, through determining the performance values of six freely accessible Web-tools (ADMETlab, AdmetSAR2.0, PgpRules, pkCSM, SwissADME and vNN-ADMET), computationally predicting P-gp-mediated transport. Using an external test set of 231 marketed drugs, approved over the 2010-2020 period by the US Food and Drug Administration and fully in vitro characterized for their P-gp substrate status, various performance parameters (including sensitivity, specificity, accuracy, Matthews correlation coefficient and area under the receiver operating characteristics curve) were determined. They were found to rather poorly meet criteria commonly required for acceptable prediction, whatever the Web-tools were used alone or in combination. Predictions of being P-gp substrate or non-substrate by these online in silico methods may therefore be considered with caution.

Automated and enabling technologies for medicinal chemistry

Having always been driven by the need to get new treatments to patients as quickly as possible, drug discovery is a constantly evolving process. This chapter will review how medicinal chemistry was established, how it has changed over the years due to the emergence of new enabling technologies, and how early advances in synthesis, purification and analysis, have provided the foundations upon which the current automated and enabling technologies are built. Looking beyond the established technologies, this chapter will also consider technologies that are now emerging, and their impact on the future of drug discovery and the role of medicinal chemists.

Could graph neural networks learn better molecular representation for drug discovery? A comparison study of descriptor-based and graph-based models

Graph neural networks (GNN) has been considered as an attractive modelling method for molecular property prediction, and numerous studies have shown that GNN could yield more promising results than traditional descriptor-based methods. In this study, based on 11 public datasets covering various property endpoints, the predictive capacity and computational efficiency of the prediction models developed by eight machine learning (ML) algorithms, including four descriptor-based models (SVM, XGBoost, RF and DNN) and four graph-based models (GCN, GAT, MPNN and Attentive FP), were extensively tested and compared. The results demonstrate that on average the descriptor-based models outperform the graph-based models in terms of prediction accuracy and computational efficiency. SVM generally achieves the best predictions for the regression tasks. Both RF and XGBoost can achieve reliable predictions for the classification tasks, and some of the graph-based models, such as Attentive FP and GCN, can yield outstanding performance for a fraction of larger or multi-task datasets. In terms of computational cost, XGBoost and RF are the two most efficient algorithms and only need a few seconds to train a model even for a large dataset. The model interpretations by the SHAP method can effectively explore the established domain knowledge for the descriptor-based models. Finally, we explored use of these models for virtual screening (VS) towards HIV and demonstrated that different ML algorithms offer diverse VS profiles. All in all, we believe that the off-the-shelf descriptor-based models still can be directly employed to accurately predict various chemical endpoints with excellent computability and interpretability.

Development of a Hierarchical Support Vector Regression-Based In Silico Model for Caco-2 Permeability

Drug absorption is one of the critical factors that should be taken into account in the process of drug discovery and development. The human colon carcinoma cell layer (Caco-2) model has been frequently used as a surrogate to preliminarily investigate the intestinal absorption. In this study, a quantitative structure–activity relationship (QSAR) model was generated using the innovative machine learning-based hierarchical support vector regression (HSVR) scheme to depict the exceedingly confounding passive diffusion and transporter-mediated active transport. The HSVR model displayed good agreement with the experimental values of the training samples, test samples, and outlier samples. The predictivity of HSVR was further validated by a mock test and verified by various stringent statistical criteria. Consequently, this HSVR model can be employed to forecast the Caco-2 permeability to assist drug discovery and development.

Artificial intelligence in the early stages of drug discovery

Although the use of computational methods within the pharmaceutical industry is well established, there is an urgent need for new approaches that can improve and optimize the pipeline of drug discovery and development. In spite of the fact that there is no unique solution for this need for innovation, there has recently been a strong interest in the use of Artificial Intelligence for this purpose. As a matter of fact, not only there have been major contributions from the scientific community in this respect, but there has also been a growing partnership between the pharmaceutical industry and Artificial Intelligence companies. Beyond these contributions and efforts there is an underlying question, which we intend to discuss in this review: can the intrinsic difficulties within the drug discovery process be overcome with the implementation of Artificial Intelligence? While this is an open question, in this work we will focus on the advantages that these algorithms provide over the traditional methods in the context of early drug discovery.

Bioactivity Comparison across Multiple Machine Learning Algorithms Using over 5000 Datasets for Drug Discovery

Machine learning methods are attracting considerable attention from the pharmaceutical industry for use in drug discovery and applications beyond. In recent studies, we and others have applied multiple machine learning algorithms and modeling metrics and, in some cases, compared molecular descriptors to build models for individual targets or properties on a relatively small scale. Several research groups have used large numbers of datasets from public databases such as ChEMBL in order to evaluate machine learning methods of interest to them. The largest of these types of studies used on the order of 1400 datasets. We have now extracted well over 5000 datasets from CHEMBL for use with the ECFP6 fingerprint and in comparison of our proprietary software Assay Central with random forest, k-nearest neighbors, support vector classification, naïve Bayesian, AdaBoosted decision trees, and deep neural networks (three layers). Model performance was assessed using an array of fivefold cross-validation metrics including area-under-the-curve, F1 score, Cohen's kappa, and Matthews correlation coefficient. Based on ranked normalized scores for the metrics or datasets, all methods appeared comparable, while the distance from the top indicated that Assay Central and support vector classification were comparable. Unlike prior studies which have placed considerable emphasis on deep neural networks (deep learning), no advantage was seen in this case. If anything, Assay Central may have been at a slight advantage as the activity cutoff for each of the over 5000 datasets representing over 570,000 unique compounds was based on Assay Central performance, although support vector classification seems to be a strong competitor. We also applied Assay Central to perform prospective predictions for the toxicity targets PXR and hERG to further validate these models. This work appears to be the largest scale comparison of these machine learning algorithms to date. Future studies will likely evaluate additional databases, descriptors, and machine learning algorithms and further refine the methods for evaluating and comparing such models.

The Intestinal Efflux Transporter Inhibition Activity of Xanthones from Mangosteen Pericarp: An In Silico, In Vitro and Ex Vivo Approach

The capacity of α-mangostin (α-MG) and β-mangostin (β-MG) from mangosteen pericarp on P-glycoprotein (Pgp) in silico, in vitro, and ex vivo was investigated in this study. Screening with the ADMET Predictor™ program predicted the two compounds to be both a Pgp inhibitor and Pgp substrate. The compounds tended to interact with Pgp and inhibit Pgp ATPase activity. Additionally, bidirectional transport on Caco-2 cell monolayers demonstrated a significantly lower efflux ratio than that of the control (α-(44.68) and β-(46.08) MG versus the control (66.26); p < 0.05) indicating an inhibitory effect on Pgp activity. Test compounds additionally revealed a downregulation of MDR1 mRNA expression. Moreover, an ex vivo absorptive transport in everted mouse ileum confirmed the previous results that α-MG had a Pgp affinity inhibitor, leading to an increase in absorption of the Pgp substrate in the serosal side. In conclusion, α- and β-MG have the capability to inhibit Pgp and they also alter Pgp expression, which makes them possible candidates for reducing multidrug resistance. Additionally, they influence the bioavailability and transport of Pgp substrate drugs.

P-Glycoprotein: One Mechanism, Many Tasks and the Consequences for Pharmacotherapy of Cancers

P-glycoprotein or multidrug resistance protein (MDR1) is an adenosine triphosphate (ATP) binding cassette transporter (ABCB1) intensely investigated because it is an obstacle to successful pharmacotherapy of cancers. P-glycoprotein prevents cellular uptake of a large number of structurally and functionally diverse compounds, including most cancer therapeutics and in this way causes multidrug resistance (MDR). To overcome MDR, and thus improve cancer treatment, an understanding of P-glycoprotein inhibition at the molecular level is required. With this goal in mind, we propose rules that predict whether a compound is a modulator, substrate, inhibitor, or inducer of P-glycoprotein. This new set of rules is derived from a quantitative analysis of the drug binding and transport properties of P-glycoprotein. We further discuss the role of P-glycoprotein in immune surveillance and cell metabolism. Finally, the predictive power of the proposed rules is demonstrated with a set of FDA approved drugs which have been repurposed for cancer therapy.

Integrating Machine Learning-Based Virtual Screening With Multiple Protein Structures and Bio-Assay Evaluation for Discovery of Novel GSK3β Inhibitors

Glycogen synthase kinase-3β (GSK3β) is associated with various key biological processes, and it has been considered as a critical therapeutic target for the treatment of many diseases. However, it is a big challenge to develop ATP-competition GSK3β inhibitors because of the high sequence homology with other kinases. In this work, a novel parallel virtual screening strategy based on multiple GSK3β protein structures, integrating molecular docking, complex-based pharmacophore, and naive Bayesian classification, was developed to screen a large chemical database, the 50 compounds with top-scores then underwent a luminescent kinase assay, which led to the discovery of two GSK3β inhibitor hits. The high screening enrichment rate indicates the reliability and practicability of the integrated protocol. Finally, molecular docking and molecular dynamics simulation were employed to investigate the binding modes of the GSK3β inhibitors, and some “hot residues” critical to GSK3β affinity were highlighted. The present study may provide some valuable guidance for the development of novel GSK3β inhibitors.

Binding Site Interactions of Modulators of Breast Cancer Resistance Protein, Multidrug Resistance-Associated Protein 2, and P-Glycoprotein Activity

ATP-binding cassette (ABC)-transporters protect tissues by pumping their substrates out of the cells in many physiological barriers, such as the blood–brain barrier, intestine, liver, and kidney. These substrates include various endogenous metabolites, but, in addition, ABC transporters recognize a wide range of compounds, therefore affecting the disposition and elimination of clinically used drugs and their metabolites. Although numerous ABC-transporter inhibitors are known, the underlying mechanism of inhibition is not well characterized. The aim of this study is to deepen our understanding of transporter inhibition by studying the molecular basis of ligand recognition. In the current work, we compared the effect of 44 compounds on the active transport mediated by three ABC transporters: breast cancer resistance protein (BCRP and ABCG2), multidrug-resistance associated protein (MRP2 and ABCC2), and P-glycoprotein (P-gp and ABCB1). Eight compounds were strong inhibitors of all three transporters, while the activity of 36 compounds was transporter-specific. Of the tested compounds, 39, 25, and 11 were considered as strong inhibitors, while 1, 4, and 11 compounds were inactive against BCRP, MRP2, and P-gp, respectively. In addition, six transport-enhancing stimulators were observed for P-gp. In order to understand the observed selectivity, we compared the surface properties of binding cavities in the transporters and performed structure–activity analysis and computational docking of the compounds to known binding sites in the transmembrane domains and nucleotide-binding domains. Based on the results, the studied compounds are more likely to interact with the transmembrane domain than the nucleotide-binding domain. Additionally, the surface properties of the substrate binding site in the transmembrane domains of the three transporters were in line with the observed selectivity. Because of the high activity toward BCRP, we lacked the dynamic range needed to draw conclusions on favorable interactions; however, we identified amino acids in both P-gp and MRP2 that appear to be important for ligand recognition.

ADMET evaluation in drug discovery. 20. Prediction of breast cancer resistance protein inhibition through machine learning

Breast cancer resistance protein (BCRP/ABCG2), an ATP-binding cassette (ABC) efflux transporter, plays a critical role in multi-drug resistance (MDR) to anti-cancer drugs and drug–drug interactions. The prediction of BCRP inhibition can facilitate evaluating potential drug resistance and drug–drug interactions in early stage of drug discovery. Here we reported a structurally diverse dataset consisting of 1098 BCRP inhibitors and 1701 non-inhibitors. Analysis of various physicochemical properties illustrates that BCRP inhibitors are more hydrophobic and aromatic than non-inhibitors. We then developed a series of quantitative structure–activity relationship (QSAR) models to discriminate between BCRP inhibitors and non-inhibitors. The optimal feature subset was determined by a wrapper feature selection method named rfSA (simulated annealing algorithm coupled with random forest), and the classification models were established by using seven machine learning approaches based on the optimal feature subset, including a deep learning method, two ensemble learning methods, and four classical machine learning methods. The statistical results demonstrated that three methods, including support vector machine (SVM), deep neural networks (DNN) and extreme gradient boosting (XGBoost), outperformed the others, and the SVM classifier yielded the best predictions (MCC = 0.812 and AUC = 0.958 for the test set). Then, a perturbation-based model-agnostic method was used to interpret our models and analyze the representative features for different models. The application domain analysis demonstrated the prediction reliability of our models. Moreover, the important structural fragments related to BCRP inhibition were identified by the information gain (IG) method along with the frequency analysis. In conclusion, we believe that the classification models developed in this study can be regarded as simple and accurate tools to distinguish BCRP inhibitors from non-inhibitors in drug design and discovery pipelines.

Bayesian machine learning to discover Bruton's tyrosine kinase inhibitors

Bruton's tyrosine kinase (BTK) has a crucial role in multiple cell signaling pathways including B-cell antigen receptor (BCR) and Fc receptor (FcR) signaling cascades, which has attracted much attention to find BTK inhibitors to treat autoimmune diseases. In this work, we constructed a Bayesian classification model for virtually seeking novel BTK inhibitors, which showed good performance in terms of screening efficiency and accuracy. Through searching for several chemical libraries including Chembl_17 (1,317,484 compounds), Chembridge (103,473 compounds), and Chemdiv (700,000 compounds) using this model followed by molecular docking and activity prediction, 52 compounds with novel scaffolds were acknowledged as potential BTK inhibitors, which could be promising starting points for further exploration. This study also provided a guide to construct an efficient and effective protocol for virtual screening by integrating machine learning methods.

Recent Advances in Developing K-Ras Plasma Membrane Localization Inhibitors

The Ras proteins play an important role in cell growth, differentiation, proliferation and survival by regulating diverse signaling pathways. Oncogenic mutant K-Ras is the most frequently mutated class of Ras superfamily that is highly prevalent in many human cancers. Despite intensive efforts to combat various K-Ras-mutant-driven cancers, no effective K-Ras-specific inhibitors have yet been approved for clinical use to date. Since K-Ras proteins must be associated to the plasma membrane for their function, targeting K-Ras plasma membrane localization represents a logical and potentially tractable therapeutic approach. Here, we summarize the recent advances in the development of K-Ras plasma membrane localization inhibitors including natural product-based inhibitors achieved from high throughput screening, fragment-based drug design, virtual screening, and drug repurposing as well as hit-to-lead optimizations.

STS-NLSP: A Network-Based Label Space Partition Method for Predicting the Specificity of Membrane Transporter Substrates Using a Hybrid Feature of Structural and Semantic Similarity

Membrane transport proteins play crucial roles in the pharmacokinetics of substrate drugs, the drug resistance in cancer and are vital to the process of drug discovery, development and anti-cancer therapeutics. However, experimental methods to profile a substrate drug against a panel of transporters to determine its specificity are labor intensive and time consuming. In this article, we aim to develop an in silico multi-label classification approach to predict whether a substrate can specifically recognize one of the 13 categories of drug transporters ranging from ATP-binding cassette to solute carrier families using both structural fingerprints and chemical ontologies information of substrates. The data-driven network-based label space partition (NLSP) method was utilized to construct the model based on a hybrid of similarity-based feature by the integration of 2D fingerprint and semantic similarity. This method builds predictors for each label cluster (possibly intersecting) detected by community detection algorithms and takes union of label sets for a compound as final prediction. NLSP lies into the ensembles of multi-label classifier category in multi-label learning field. We utilized Cramér's V statistics to quantify the label correlations and depicted them via a heatmap. The jackknife tests and iterative stratification based cross-validation method were adopted on a benchmark dataset to evaluate the prediction performance of the proposed models both in multi-label and label-wise manner. Compared with other powerful multi-label methods, ML-kNN, MTSVM, and RAkELd, our multi-label classification model of NLPS-RF (random forest-based NLSP) has proven to be a feasible and effective model, and performed satisfactorily in the predictive task of transporter-substrate specificity. The idea behind NLSP method is intriguing and the power of NLSP remains to be explored for the multi-label learning problems in bioinformatics. The benchmark dataset, intermediate results and python code which can fully reproduce our experiments and results are available at https://github.com/dqwei-lab/STS.

A drug-likeness toolbox facilitates ADMET study in drug discovery

Undesirable pharmacokinetic (PK) properties or unacceptable toxicity are the main causes of the failure of drug candidates at the clinical trial stage. Since the concept of drug-likeness was first proposed, it has become an important consideration in the selection of compounds with desirable bioavailability during the early phases of drug discovery. Over the past decade, online resources have effectively facilitated drug-likeness studies in an economical and time-efficient manner. Here, we provide a comprehensive summary and comparison of current accessible online resources, in terms of their key features, application fields, and performance for in silico drug-likeness studies. We hope that the assembled toolbox will provide useful guidance to facilitate future in silico drug-likeness research.

Multiclass Classifier for P-Glycoprotein Substrates, Inhibitors, and Non-Active Compounds

P-glycoprotein (P-gp) is a transmembrane protein that actively transports a wide variety of chemically diverse compounds out of the cell. It is highly associated with the ADMET (absorption, distribution, metabolism, excretion and toxicity) properties of drugs/drug candidates and contributes to decreasing toxicity by eliminating compounds from cells, thereby preventing intracellular accumulation. Therefore, in the drug discovery and toxicological assessment process it is advisable to pay attention to whether a compound under development could be transported by P-gp or not. In this study, an in silico multiclass classification model capable of predicting the probability of a compound to interact with P-gp was developed using a counter-propagation artificial neural network (CP ANN) based on a set of 2D molecular descriptors, as well as an extensive dataset of 2512 compounds (1178 P-gp inhibitors, 477 P-gp substrates and 857 P-gp non-active compounds). The model provided a good classification performance, producing non error rate (NER) values of 0.93 for the training set and 0.85 for the test set, while the average precision (AvPr) was 0.93 for the training set and 0.87 for the test set. An external validation set of 385 compounds was used to challenge the model’s performance. On the external validation set the NER and AvPr values were 0.70 for both indices. We believe that this in silico classifier could be effectively used as a reliable virtual screening tool for identifying potential P-gp ligands.

Antitrypanosomal Activity of Acetogenins Isolated from the Seeds of Porcelia macrocarpa Is Associated with Alterations in Both Plasma Membrane Electric Potential and Mitochondrial Membrane Potential

As part of a drug discovery program aimed at the identification of anti- Trypanosoma cruzi metabolites from Brazilian flora, four acetogenins (1-4) were isolated from the seeds of Porcelia macrocarpa and were identified by NMR spectroscopy and HRESIMS. The new compounds 1 and 2 displayed activity against the trypomastigote (IC₅₀ = 0.4 and 3.6 μM) and amastigote (IC₅₀ = 23.0 and 27.7 μM) forms. The structurally related known compound 3 showed less potency to the amastigotes, with an IC₅₀ value of 58 μM, while the known compound 4 was inactive. To evaluate the potential mechanisms for parasite death, parameters were evaluated by fluorometric assays: (i) plasma membrane permeability, (ii) plasma membrane electric potential (ΔΨ_p), (iii) reactive oxygen species production, and (iv) mitochondrial membrane potential (ΔΨ_m). The results obtained indicated that compounds 1 and 2 depolarize plasma membranes, affecting ΔΨ_p and ΔΨ_m and contributing to the observed cellular damage and disturbing the bioenergetic system. In silico studies of pharmacokinetics and toxicity (ADMET) properties predicted that all compounds were nonmutagenic, noncarcinogenic, nongenotoxic, and weak hERG blockers. Additionally, none of the isolated acetogenins 1-4 were predicted as pan-assay interference compounds.

Characterizing ABC-Transporter Substrate-Likeness Using a Clean-Slate Genetic Background

Mutations in ATP Binding Cassette (ABC)-transporter genes can have major effects on the bioavailability and toxicity of the drugs that are ABC-transporter substrates. Consequently, methods to predict if a drug is an ABC-transporter substrate are useful for drug development. Such methods traditionally relied on literature curated collections of ABC-transporter dependent membrane transfer assays. Here, we used a single large-scale dataset of 376 drugs with relative efficacy on an engineered yeast strain with all ABC-transporter genes deleted (ABC-16), to explore the relationship between a drug’s chemical structure and ABC-transporter substrate-likeness. We represented a drug’s chemical structure by an array of substructure keys and explored several machine learning methods to predict the drug’s efficacy in an ABC-16 yeast strain. Gradient-Boosted Random Forest models outperformed all other methods with an AUC of 0.723. We prospectively validated the model using new experimental data and found significant agreement with predictions. Our analysis expands the previously reported chemical substructures associated with ABC-transporter substrates and provides an alternative means to investigate ABC-transporter substrate-likeness.

Development of Simplified in Vitro P-Glycoprotein Substrate Assay and in Silico Prediction Models To Evaluate Transport Potential of P-Glycoprotein

For efficient drug discovery and screening, it is necessary to simplify P-glycoprotein (P-gp) substrate assays and to provide in silico models that predict the transport potential of P-gp. In this study, we developed a simplified in vitro screening method to evaluate P-gp substrates by unidirectional membrane transport in P-gp-overexpressing cells. The unidirectional flux ratio positively correlated with parameters of the conventional bidirectional P-gp substrate assay ( R² = 0.941) and in vivo K_p,brain ratio (mdr1a/1b KO/WT) in mice ( R² = 0.800). Our in vitro P-gp substrate assay had high reproducibility and required approximately half the labor of the conventional method. We also constructed regression models to predict the value of P-gp-mediated flux and three-class classification models to predict P-gp substrate potential (low-, medium-, and high-potential) using 2397 data entries with the largest data set collected under the same experimental conditions. Most compounds in the test set fell within two- and three-fold errors in the random forest regression model (71.3 and 88.5%, respectively). Furthermore, the random forest three-class classification model showed a high balanced accuracy of 0.821 and precision of 0.761 for the low-potential classes in the test set. We concluded that the simplified in vitro P-gp substrate assay was suitable for compound screening in the early stages of drug discovery and that the in silico regression model and three-class classification model using only chemical structure information could identify the transport potential of compounds including P-gp-mediated flux ratios. Our proposed method is expected to be a practical tool to optimize effective central nervous system (CNS) drugs, to avoid CNS side effects, and to improve intestinal absorption.

PgpRules: a decision tree based prediction server for P-glycoprotein substrates and inhibitors

Summary

P-glycoprotein (P-gp) is a member of ABC transporter family that actively pumps xenobiotics out of cells to protect organisms from toxic compounds. P-gp substrates can be easily pumped out of the cells to reduce their absorption; conversely P-gp inhibitors can reduce such pumping activity. Hence, it is crucial to know if a drug is a P-gp substrate or inhibitor in view of pharmacokinetics. Here we present PgpRules, an online P-gp substrate and P-gp inhibitor prediction server with ruled-sets. The two models were built using classification and regression tree algorithm. For each compound uploaded, PgpRules not only predicts whether the compound is a P-gp substrate or a P-gp inhibitor, but also provides the rules containing chemical structural features for further structural optimization.

Availability and implementation

PgpRules is freely accessible at https://pgprules.cmdm.tw/.

Supplementary information

Supplementary data are available at Bioinformatics online.

An Integrated Transfer Learning and Multitask Learning Approach for Pharmacokinetic Parameter Prediction

BACKGROUND

Pharmacokinetic evaluation is one of the key processes in drug discovery and development. However, current absorption, distribution, metabolism, and excretion prediction models still have limited accuracy.

AIM

This study aims to construct an integrated transfer learning and multitask learning approach for developing quantitative structure-activity relationship models to predict four human pharmacokinetic parameters.

METHODS

A pharmacokinetic data set included 1104 U.S. FDA approved small molecule drugs. The data set included four human pharmacokinetic parameter subsets (oral bioavailability, plasma protein binding rate, apparent volume of distribution at steady-state, and elimination half-life). The pretrained model was trained on over 30 million bioactivity data entries. An integrated transfer learning and multitask learning approach was established to enhance the model generalization.

RESULTS

The pharmacokinetic data set was split into three parts (60:20:20) for training, validation, and testing by the improved maximum dissimilarity algorithm with the representative initial set selection algorithm and the weighted distance function. The multitask learning techniques enhanced the model predictive ability. The integrated transfer learning and multitask learning model demonstrated the best accuracies, because deep neural networks have the general feature extraction ability; transfer learning and multitask learning improve the model generalization.

CONCLUSIONS

The integrated transfer learning and multitask learning approach with the improved data set splitting algorithm was first introduced to predict the pharmacokinetic parameters. This method can be further employed in drug discovery and development.

Analogues of the Lignan Pinoresinol as Novel Lead Compounds for P-glycoprotein (P-gp) Inhibitors

To find novel P-gp-inhibitors, a library of pregnane X receptor (PXR) ligands and the ZINC DrugsNow library were superimposed on the P-gp inhibitor (+)-pinoresinol (1) used as a query for a three-dimensional similarity search. After determining the TanimotoCombo index of similarity with 1, eight compounds from the PXR library and two ZINC compounds were selected for biological evaluation. The P-gp inhibition study showed that compounds 7, 8, and 9 successfully increased intracellular doxorubicin (DOX) accumulation in the P-gp overexpressed Lucena 1 cells from 25, 12.5, and 6.25 μM, respectively. Among a series of analogues of 9, compounds 26-30 were shown to be active, with 26 and 27 causing a significant increase in DOX accumulation from 1.56 μM and rendering Lucena 1 sensitive to DOX from 1.56 and 0.78 μM, respectively. Molecular modeling studies showed that both compounds bind to the P-gp at transmembrane helices (TMH) 4, 5, and 6, with 27 also showing contacts with TMH 3.

Theoretical Prediction of the Complex P-Glycoprotein Substrate Efflux Based on the Novel Hierarchical Support Vector Regression Scheme

P-glycoprotein (P-gp), a membrane-bound transporter, can eliminate xenobiotics by transporting them out of the cells or blood⁻brain barrier (BBB) at the expense of ATP hydrolysis. Thus, P-gp mediated efflux plays a pivotal role in altering the absorption and disposition of a wide range of substrates. Nevertheless, the mechanism of P-gp substrate efflux is rather complex since it can take place through active transport and passive permeability in addition to multiple P-gp substrate binding sites. A nonlinear quantitative structure⁻activity relationship (QSAR) model was developed in this study using the novel machine learning-based hierarchical support vector regression (HSVR) scheme to explore the perplexing relationships between descriptors and efflux ratio. The predictions by HSVR were found to be in good agreement with the observed values for the molecules in the training set (n = 50, r² = 0.96, qCV2 = 0.94, RMSE = 0.10, s = 0.10) and test set (n = 13, q² = 0.80⁻0.87, RMSE = 0.21, s = 0.22). When subjected to a variety of statistical validations, the developed HSVR model consistently met the most stringent criteria. A mock test also asserted the predictivity of HSVR. Consequently, this HSVR model can be adopted to facilitate drug discovery and development.

MetStabOn-Online Platform for Metabolic Stability Predictions

Metabolic stability is an important parameter to be optimized during the complex process of designing new active compounds. Tuning this parameter with the simultaneous maintenance of a desired compound's activity is not an easy task due to the extreme complexity of metabolic pathways in living organisms. In this study, the platform for in silico qualitative evaluation of metabolic stability, expressed as half-lifetime and clearance was developed. The platform is based on the application of machine learning methods and separate models for human, rat and mouse data were constructed. The compounds' evaluation is qualitative and two types of experiments can be performed-regression, which is when the compound is assigned to one of the metabolic stability classes (low, medium, high) on the basis of numerical value of the predicted half-lifetime, and classification, in which the molecule is directly assessed as low, medium or high stability. The results show that the models have good predictive power, with accuracy values over 0.7 for all cases, for Sequential Minimal Optimization (SMO), k-nearest neighbor (IBk) and Random Forest algorithms. Additionally, for each of the analyzed compounds, 10 of the most similar structures from the training set (in terms of Tanimoto metric similarity) are identified and made available for download as separate files for more detailed manual inspection. The predictive power of the models was confronted with the external dataset, containing metabolic stability assessment via the GUSAR software, leading to good consistency of results for SMOreg and Naïve Bayes (~0.8 on average). The tool is available online.

Benchmark Study Based on 2P2IDB to Gain Insights into the Discovery of Small-Molecule PPI Inhibitors

Protein-protein interactions (PPIs) have been regarded as novel and highly promising drug targets in drug discovery. Numerous new experimental techniques and computational approaches have been developed to assist the design of PPI modulators during the past two decades. However, identification and optimization of small-molecule inhibitors targeting PPIs is still a particularly challenging task due to the "undruggable" profiles of PPI interfaces. Nowadays, in silico screening, especially docking-based virtual screening, has emerged as an effective method to complement experimental high-throughput screening in identifying novel and potent small-molecule PPI inhibitors. Here, on the basis of the 2P2I_DB database, we explored the structural features of the known small-molecule PPI inhibitors and analyzed the characteristics of the PPI binding pockets. More importantly, we evaluated the sampling power and screening power of six popular docking programs for PPI targets. Our results indicate that the chlorinated conjugate group and amidelike linkage are two types of privileged fragments of PPI inhibitors; the average druggability of the binding sites of the PPI targets in 2P2IDB is slightly worse than that of traditional ones; both academic and commercial docking programs exhibit an acceptable accuracy on pose prediction for PPI inhibitors, but their screening powers for identifying PPI inhibitors are still not satisfactory. It is expected that our work can provide valuable guidance on the construction of PPI-focused library, the determination of druggable PPI binding pocket, and the selection of docking program for the screening of small-molecule PPI inhibitors.