High-throughput screening and Bayesian machine learning for copper-dependent inhibitors of Staphylococcus aureus.

One potential source of new antibacterials is through probing existing chemical libraries for copper-dependent inhibitors (CDIs), i.e., molecules with antibiotic activity only in the presence of copper. Recently, our group demonstrated that previously unknown staphylococcal CDIs were frequently present in a small pilot screen. Here, we report the outcome of a larger industrial anti-staphylococcal screen consisting of 40 771 compounds assayed in parallel, both in standard and in copper-supplemented media. Ultimately, 483 had confirmed copper-dependent IC50 values under 50 µM. Sphere-exclusion clustering revealed that these hits were largely dominated by sulfur-containing motifs, including benzimidazole-2-thiones, thiadiazines, thiazoline formamides, triazino-benzimidazoles, and pyridinyl thieno-pyrimidines. Structure-activity relationship analysis of the pyridinyl thieno-pyrimidines generated multiple improved CDIs, with activity likely dependent on ligand/ion coordination. Molecular fingerprint-based Bayesian classification models were built using Discovery Studio and Assay Central, a new platform for sharing and distributing cheminformatic models in a portable format, based on open-source tools. Finally, we used the latter model to evaluate a library of FDA-approved drugs for copper-dependent activity in silico. Two anti-helminths, albendazole and thiabendazole, scored highly and are known to coordinate copper ions, further validating the model's applicability.

High Throughput and Computational Repurposing for Neglected Diseases.

PURPOSE: Neglected tropical diseases (NTDs) represent are a heterogeneous group of communicable diseases that are found within the poorest populations of the world. There are 23 NTDs that have been prioritized by the World Health Organization, which are endemic in 149 countries and affect more than 1.4 billion people, costing these developing economies billions of dollars annually. The NTDs result from four different causative pathogens: protozoa, bacteria, helminth and virus. The majority of the diseases lack effective treatments. Therefore, new therapeutics for NTDs are desperately needed. METHODS: We describe various high throughput screening and computational approaches that have been performed in recent years. We have collated the molecules identified in these studies and calculated molecular properties. RESULTS: Numerous global repurposing efforts have yielded some promising compounds for various neglected tropical diseases. These compounds when analyzed as one would expect appear drug-like. Several large datasets are also now in the public domain and this enables machine learning models to be constructed that then facilitate the discovery of new molecules for these pathogens. CONCLUSIONS: In the space of a few years many groups have either performed experimental or computational repurposing high throughput screens against neglected diseases. These have identified compounds which in many cases are already approved drugs. Such approaches perhaps offer a more efficient way to develop treatments which are generally not a focus for global pharmaceutical companies because of the economics or the lack of a viable market. Other diseases could perhaps benefit from these repurposing approaches.

Modeling error in experimental assays using the bootstrap principle: understanding discrepancies between assays using different dispensing technologies.

All experimental assay data contains error, but the magnitude, type, and primary origin of this error is often not obvious. Here, we describe a simple set of assay modeling techniques based on the bootstrap principle that allow sources of error and bias to be simulated and propagated into assay results. We demonstrate how deceptively simple operations--such as the creation of a dilution series with a robotic liquid handler--can significantly amplify imprecision and even contribute substantially to bias. To illustrate these techniques, we review an example of how the choice of dispensing technology can impact assay measurements, and show how large contributions to discrepancies between assays can be easily understood and potentially corrected for. These simple modeling techniques--illustrated with an accompanying IPython notebook--can allow modelers to understand the expected error and bias in experimental datasets, and even help experimentalists design assays to more effectively reach accuracy and imprecision goals.

Combining Metabolite-Based Pharmacophores with Bayesian Machine Learning Models for Mycobacterium tuberculosis Drug Discovery.

Integrated computational approaches for Mycobacterium tuberculosis (Mtb) are useful to identify new molecules that could lead to future tuberculosis (TB) drugs. Our approach uses information derived from the TBCyc pathway and genome database, the Collaborative Drug Discovery TB database combined with 3D pharmacophores and dual event Bayesian models of whole-cell activity and lack of cytotoxicity. We have prioritized a large number of molecules that may act as mimics of substrates and metabolites in the TB metabolome. We computationally searched over 200,000 commercial molecules using 66 pharmacophores based on substrates and metabolites from Mtb and further filtering with Bayesian models. We ultimately tested 110 compounds in vitro that resulted in two compounds of interest, BAS 04912643 and BAS 00623753 (MIC of 2.5 and 5 µg/mL, respectively). These molecules were used as a starting point for hit-to-lead optimization. The most promising class proved to be the quinoxaline di-N-oxides, evidenced by transcriptional profiling to induce mRNA level perturbations most closely resembling known protonophores. One of these, SRI58 exhibited an MIC = 1.25 µg/mL versus Mtb and a CC50 in Vero cells of >40 µg/mL, while featuring fair Caco-2 A-B permeability (2.3 x 10-6 cm/s), kinetic solubility (125 µM at pH 7.4 in PBS) and mouse metabolic stability (63.6% remaining after 1 h incubation with mouse liver microsomes). Despite demonstration of how a combined bioinformatics/cheminformatics approach afforded a small molecule with promising in vitro profiles, we found that SRI58 did not exhibit quantifiable blood levels in mice.

Thiophenecarboxamide Derivatives Activated by EthA Kill Mycobacterium tuberculosis by Inhibiting the CTP Synthetase PyrG.

To combat the emergence of drug-resistant strains of Mycobacterium tuberculosis, new antitubercular agents and novel drug targets are needed. Phenotypic screening of a library of 594 hit compounds uncovered two leads that were active against M. tuberculosis in its replicating, non-replicating, and intracellular states: compounds 7947882 (5-methyl-N-(4-nitrophenyl)thiophene-2-carboxamide) and 7904688 (3-phenyl-N-[(4-piperidin-1-ylphenyl)carbamothioyl]propanamide). Mutants resistant to both compounds harbored mutations in ethA (rv3854c), the gene encoding the monooxygenase EthA, and/or in pyrG (rv1699) coding for the CTP synthetase, PyrG. Biochemical investigations demonstrated that EthA is responsible for the activation of the compounds, and by mass spectrometry we identified the active metabolite of 7947882, which directly inhibits PyrG activity. Metabolomic studies revealed that pharmacological inhibition of PyrG strongly perturbs DNA and RNA biosynthesis, and other metabolic processes requiring nucleotides. Finally, the crystal structure of PyrG was solved, paving the way for rational drug design with this newly validated drug target.

A virtual screen discovers novel, fragment-sized inhibitors of Mycobacterium tuberculosis InhA.

Isoniazid (INH) is usually administered to treat latent Mycobacterium tuberculosis (Mtb) infections and is used in combination therapy to treat active tuberculosis (TB). Unfortunately, resistance to this drug is hampering its clinical effectiveness. INH is a prodrug that must be activated by Mtb catalase-peroxidase (KatG) before it can inhibit InhA (Mtb enoyl-acyl-carrier-protein reductase). Isoniazid-resistant cases of TB found in clinical settings usually involve mutations in or deletion of katG, which abrogate INH activation. Compounds that inhibit InhA without requiring prior activation by KatG would not be affected by this resistance mechanism and hence would display continued potency against these drug-resistant isolates of Mtb. Virtual screening experiments versus InhA in the GO Fight Against Malaria (GO FAM) project were designed to discover new scaffolds that display base-stacking interactions with the NAD cofactor. GO FAM experiments included targets from other pathogens, including Mtb, when they had structural similarity to a malaria target. Eight of the 16 soluble compounds identified by docking against InhA plus visual inspection were modest inhibitors and did not require prior activation by KatG. The best two inhibitors discovered are both fragment-sized compounds and displayed Ki values of 54 and 59 µM, respectively. Importantly, the novel inhibitors discovered have low structural similarity to known InhA inhibitors and thus help expand the number of chemotypes on which future medicinal chemistry efforts can be focused. These new fragment hits could eventually help advance the fight against INH-resistant Mtb strains, which pose a significant global health threat.

Quantitative NTCP pharmacophore and lack of association between DILI and NTCP Inhibition.

The human sodium taurocholate cotransporting polypeptide (NTCP) is a hepatic bile acid transporter. Inhibition of NTCP uptake may potentially also prevent hepatitis B virus (HBV) infection. The first objective was to develop a quantitative pharmacophore for NTCP inhibition. Recent studies showed that hepatotoxic drugs could inhibit bile acid uptake into hepatocytes, without inhibiting canalicular efflux, and cause bile acid elevation in plasma. Hence, a second objective was to examine whether NTCP inhibition is associated with drug induced liver injury (DILI). Twenty-seven drugs from our previous study were used as the training set to develop a quantitative pharmacophore. From secondary screening from a drug database, six retrieved drugs and three drugs not retrieved by the model were tested for NTCP inhibition. Tertiary screening involved drugs known to cause DILI and not cause DILI. Overall, ninety-four drugs were assessed for hepatotoxicity and were assessed relative to NTCP inhibition. The quantitative pharmacophore possessed one hydrogen bond acceptor, one hydrogen bond donor, a hydrophobic feature, and excluded volumes. From 94 drugs, NTCP inhibitors and non-inhibitors were approximately equally distributed across the drugs of most DILI concern, less DILI concern, and no DILI concern, indicating no relationship between NTCP inhibition and DILI risk. Hence, an approach to treat HBV via NTCP inhibition is not expected to be associated with DILI.

Are bigger data sets better for machine learning? Fusing single-point and dual-event dose response data for Mycobacterium tuberculosis.

Tuberculosis is a major, neglected disease for which the quest to find new treatments continues. There is an abundance of data from large phenotypic screens in the public domain against Mycobacterium tuberculosis (Mtb). Since machine learning methods can learn from past data, we were interested in addressing whether more data builds better models. We now describe using Bayesian machine learning to assess whether we can improve our models by combining the large quantities of single-point data with the much smaller (higher quality) dual-event data sets, which use both dose-response data for both whole-cell antitubercular activity and Vero cell cytotoxicity. We have evaluated 12 models ranging from different single-point, dual-event dose-response, single-point and dual-event dose-response as well as combined data sets for three distinct data sets from the same laboratory. We used a fourth data set of active and inactive compounds from the same group as well as a smaller set of 177 active compounds from GlaxoSmithKline as test sets. Our data suggest combining single-point with dual-event dose-response data does not diminish the internal or external predictive ability of the models based on the receiver operator curve (ROC) for these models (internal ROC range 0.83-0.91, external ROC range 0.62-0.83) compared to the orders of magnitude smaller dual-event models (internal ROC range 0.6-0.83 and external ROC 0.54-0.83). In conclusion, models developed with 1200-5000 compounds appear to be as predictive as those generated with 25 000-350 000 molecules. Our results have implications for justifying further high-throughput screening versus focused testing based on model predictions.

Bayesian models leveraging bioactivity and cytotoxicity information for drug discovery.

Identification of unique leads represents a significant challenge in drug discovery. This hurdle is magnified in neglected diseases such as tuberculosis. We have leveraged public high-throughput screening (HTS) data to experimentally validate a virtual screening approach employing Bayesian models built with bioactivity information (single-event model) as well as bioactivity and cytotoxicity information (dual-event model). We virtually screened a commercial library and experimentally confirmed actives with hit rates exceeding typical HTS results by one to two orders of magnitude. This initial dual-event Bayesian model identified compounds with antitubercular whole-cell activity and low mammalian cell cytotoxicity from a published set of antimalarials. The most potent hit exhibits the in vitro activity and in vitro/in vivo safety profile of a drug lead. These Bayesian models offer significant economies in time and cost to drug discovery.

Ligand- and structure-based pregnane X receptor models.

The human pregnane X receptor (PXR) is a ligand dependent transcription factor that can be activated by structurally diverse agonists including steroid hormones, bile acids, herbal drugs, and prescription medications. PXR regulates the transcription of several genes involved in xenobiotic detoxification and apoptosis. Activation of PXR has the potential to initiate adverse effects by altering drug pharmacokinetics or perturbing physiological processes. Hence, more reliable prediction of PXR activators would be valuable for pharmaceutical drug discovery to avoid potential toxic effects. Ligand- and protein structure-based computational models for PXR activation have been developed in several studies. There has been limited success with structure-based modeling approaches to predict human PXR activators, which can be attributed to the large and promiscuous site of this protein. Slightly better success has been achieved with ligand-based modeling methods including quantitative structure-activity relationship (QSAR) analysis, pharmacophore modeling and machine learning that use appropriate descriptors to account for the diversity of the ligand classes that bind to PXR. These combined computational approaches using molecular shape information may assist scientists to more confidently identify PXR activators. This chapter reviews the various ligand and structure based methods undertaken to date and their results.

Identification and validation of novel human pregnane X receptor activators among prescribed drugs via ligand-based virtual screening.

Human pregnane X receptor (hPXR) plays a key role in regulating metabolism and clearance of endogenous and exogenous substances. Identification of novel hPXR activators among commercial drugs may aid in avoiding drug-drug interactions during coadministration. We applied ligand-based computational approaches for virtual screening of a commonly prescribed drug database (SCUT). Bayesian classification models were generated with a training set comprising 177 compounds using Fingerprints and 117 structural descriptors. A cell-based luciferase reporter assay was used for evaluation of chemical-mediated hPXR activation in HepG2 cells. All compounds were tested at 10 µM concentration with rifampicin and dimethyl sulfoxide as positive and negative controls, respectively. The Bayesian models showed specificity and overall prediction accuracy up to 0.92 and 0.69 for test set compounds. Screening the SCUT database with this model retrieved 105 hits and 17 compounds from the top 25 hits were chosen for in vitro testing. The reporter assay confirmed that nine drugs, i.e., fluticasone, nimodipine, nisoldipine, beclomethasone, finasteride, flunisolide, megestrol, secobarbital, and aminoglutethimide, were previously unidentified hPXR activators. Thus, the present study demonstrates that novel hPXR activators can be efficiently identified among U.S. Food and Drug Administration-approved and commonly prescribed drugs, which should lead to detection and prevention of potential drug-drug interactions.

Analysis and hit filtering of a very large library of compounds screened against Mycobacterium tuberculosis.

There is an urgent need for new drugs against tuberculosis which annually claims 1.7-1.8 million lives. One approach to identify potential leads is to screen in vitro small molecules against Mycobacterium tuberculosis (Mtb). Until recently there was no central repository to collect information on compounds screened. Consequently, it has been difficult to analyze molecular properties of compounds that inhibit the growth of Mtb in vitro. We have collected data from publically available sources on over 300 000 small molecules deposited in the Collaborative Drug Discovery TB Database. A cheminformatics analysis on these compounds indicates that inhibitors of the growth of Mtb have statistically higher mean logP, rule of 5 alerts, while also having lower HBD count, atom count and lower PSA (ChemAxon descriptors), compared to compounds that are classed as inactive. Additionally, Bayesian models for selecting Mtb active compounds were evaluated with over 100 000 compounds and, they demonstrated 10 fold enrichment over random for the top ranked 600 compounds. This represents a promising approach for finding compounds active against Mtb in whole cells screened under the same in vitro conditions. Various sets of Mtb hit molecules were also examined by various filtering rules used widely in the pharmaceutical industry to identify compounds with potentially reactive moieties. We found differences between the number of compounds flagged by these rules in Mtb datasets, malaria hits, FDA approved drugs and antibiotics. Combining these approaches may enable selection of compounds with increased probability of inhibition of whole cell Mtb activity.

When pharmaceutical companies publish large datasets: an abundance of riches or fool's gold?

The recent announcement that GlaxoSmithKline have released a huge tranche of whole-cell malaria screening data to the public domain, accompanied by a corresponding publication, raises some issues for consideration before this exemplar instance becomes a trend. We have examined the data from a high level, by studying the molecular properties, and consider the various alerts presently in use by major pharma companies. We not only acknowledge the potential value of such data but also raise the issue of the actual value of such datasets released into the public domain. We also suggest approaches that could enhance the value of such datasets to the community and theoretically offer an immediate benefit to the search for leads for other neglected diseases.

Troubleshooting computational methods in drug discovery.

Computational approaches for drug discovery such as ligand-based and structure-based methods, are increasingly seen as an efficient approach for lead discovery as well as providing insights on absorption, distribution, metabolism, excretion and toxicity (ADME/Tox). What is perhaps less well known and widely described are the limitations of the different technologies. We have therefore presented a troubleshooting approach to QSAR, homology modeling, docking as well as hybrid methods. If such computational or cheminformatics methods are to become more widely used by non-experts it is critical that such limitations are brought to the user's attention and addressed during their workflows. This could improve the quality of the models and results that are obtained.

Chemical target and pathway toxicity mechanisms defined in primary human cell systems.

INTRODUCTION: The ability to predict the health effects resulting from drug or chemical exposure has been challenging due to the complexity of human biology. Approaches that detect and discriminate a broad range of mechanisms in testing formats that are predictive and yet cost-effective are needed. METHODS: Here, we evaluated the performance of BioMAP systems, primary human cell-based disease models, as a platform for characterization of chemical toxicity mechanisms. For this we tested a set of compounds with known or well-studied mechanisms in a panel of 8 BioMAP assays relevant to human respiratory, skin, immune and vascular exposure sites. RESULTS: We evaluated the ability to detect and distinguish compounds based on mechanisms of action, comparing the BioMAP activity profiles generated in a reduced sample number format to reference database profiles derived from multiple experiments. We also studied the role of BioMAP assay panel size and concentration effects, both of which were found to contribute to the ability to discriminate chemicals and mechanisms. DISCUSSION: Compounds with diverse mechanisms, including modulators of the NFkappaB pathway, microtubule function and mitochondrial activity, could be discriminated and classified into target and pathway mechanisms in both assay formats. Certain inhibitors of mitochondrial function, such as rotenone and sodium azide, but not others, were classified with inducers of endoplasmic reticulum stress, providing insight into the toxicity mechanisms of these agents. This method may have utility in classifying novel agents with unknown modes of action according to their effects on toxicity pathways.

Precompetitive preclinical ADME/Tox data: set it free on the web to facilitate computational model building and assist drug development.

Web-based technologies coupled with a drive for improved communication between scientists have resulted in the proliferation of scientific opinion, data and knowledge at an ever-increasing rate. The increasing array of chemistry-related computer-based resources now available provides chemists with a direct path to the discovery of information, once previously accessed via library services and limited to commercial and costly resources. We propose that preclinical absorption, distribution, metabolism, excretion and toxicity data as well as pharmacokinetic properties from studies published in the literature (which use animal or human tissues in vitro or from in vivo studies) are precompetitive in nature and should be freely available on the web. This could be made possible by curating the literature and patents, data donations from pharmaceutical companies and by expanding the currently freely available ChemSpider database of over 21 million molecules with physicochemical properties. This will require linkage to PubMed, PubChem and Wikipedia as well as other frequently used public databases that are currently used, mining the full text publications to extract the pertinent experimental data. These data will need to be extracted using automated and manual methods, cleaned and then published to the ChemSpider or other database such that it will be freely available to the biomedical research and clinical communities. The value of the data being accessible will improve development of drug molecules with good ADME/Tox properties, facilitate computational model building for these properties and enable researchers to not repeat the failures of past drug discovery studies.

Computational models for drug inhibition of the human apical sodium-dependent bile acid transporter.

The human apical sodium-dependent bile acid transporter (ASBT; SLC10A2) is the primary mechanism for intestinal bile acid reabsorption. In the colon, secondary bile acids increase the risk of cancer. Therefore, drugs that inhibit ASBT have the potential to increase the risk of colon cancer. The objectives of this study were to identify FDA-approved drugs that inhibit ASBT and to derive computational models for ASBT inhibition. Inhibition was evaluated using ASBT-MDCK monolayers and taurocholate as the model substrate. Computational modeling employed a HipHop qualitative approach, a Hypogen quantitative approach, and a modified Laplacian Bayesian modeling method using 2D descriptors. Initially, 30 compounds were screened for ASBT inhibition. A qualitative pharmacophore was developed using the most potent 11 compounds and applied to search a drug database, yielding 58 hits. Additional compounds were tested, and their K(i) values were measured. A 3D-QSAR and a Bayesian model were developed using 38 molecules. The quantitative pharmacophore consisted of one hydrogen bond acceptor, three hydrophobic features, and five excluded volumes. Each model was further validated with two external test sets of 30 and 19 molecules. Validation analysis showed both models exhibited good predictability in determining whether a drug is a potent or nonpotent ASBT inhibitor. The Bayesian model correctly ranked the most active compounds. In summary, using a combined in vitro and computational approach, we found that many FDA-approved drugs from diverse classes, such as the dihydropyridine calcium channel blockers and HMG CoA-reductase inhibitors, are ASBT inhibitors.

Understanding nuclear receptors using computational methods.

Nuclear receptors (NRs) are important targets for therapeutic drugs. NRs regulate transcriptional activities through binding to ligands and interacting with several regulating proteins. Computational methods can provide insights into essential ligand-receptor and protein-protein interactions. These in turn have facilitated the discovery of novel agonists and antagonists with high affinity and specificity as well as have aided in the prediction of toxic side effects of drugs by identifying possible off-target interactions. Here, we review the application of computational methods toward several clinically important NRs (with special emphasis on PXR) and discuss their use for screening and predicting the toxic side effects of xenobiotics.

A retrospective randomized study of asthma control in the US: results of the CHARIOT study.

BACKGROUND: The third version of the National Asthma Education and Prevention Program (NAEPP) Expert Panel Report (EPR-3): Guidelines on the Diagnosis and Management of Asthma emphasizes the need to use asthma control rather than patient severity to base adjustments to treatment and ultimately improve patient outcomes. The objectives of the current study were to assess control of patients with moderate-to-severe asthma, examine the natural history of the disease, practice patterns and resource utilization in specialty community practices according to recently reviewed NAEPP guidelines. RESEARCH DESIGN AND METHODS: This analysis represents a retrospective, multicenter, randomized study of 1009 patient charts in sixty United States allergy and pulmonary medicine community practices. The proportion of patients with controlled and uncontrolled asthma over 12 months, prevalence and characteristics of atopy, past asthma history, pulmonary function, medications and treatment patterns, patient and clinical practice characteristics were analyzed. MAIN OUTCOME MEASURES: The primary outcome of interest was asthma control. RESULTS: A total of 365 male and 644 female patients with moderate-to-severe persistent asthma (mean 43.2 +/- 17.1 years) were enrolled. 81.9% of patients were uncontrolled according to recent NAEPP guidelines. Importantly, a greater percentage of patients with moderate asthma vs. severe persistent asthma were uncontrolled (p < 0.0114). Atopy was detected in 92% of patients. Patients with early onset of asthma were associated with control (p < 0.0433). Atopic symptoms, such as allergic rhinitis (p < 0.0130) and rhinosinusitis (p < 0.0476), were associated with uncontrolled asthma. Uncontrolled patients were also associated with more medications (a mean of 4.05 +/- 1.87 medications) than were controlled patients (a mean of 3.40 +/- 1.37 medications (p < 0.0001), although the temporal relationship of this association was not recorded. Limitations may have included patient and/or study site selection bias and difficulty in the process of operationalizing the definitions of control and disease severity. Since the current study only examined patients from specialty practices, the results may not be generalizable to the overall asthma population. CONCLUSIONS: Greater than 80% of asthma patients from specialty practices were uncontrolled with regard to asthma symptoms. Atopic symptoms, such as allergic rhinitis and rhinosinusitis, in addition to a greater number of medications, were associated with uncontrolled asthma. Moreover, patients designated as having asthma of moderate severity were associated with being uncontrolled more than were those with severe asthma (p < 0.0114), which suggests that the former population may not have received adequate assessment of impairment or risk, with subsequent changes in treatment for control of symptoms.

Machine learning methods and docking for predicting human pregnane X receptor activation.

The pregnane X receptor (PXR) regulates the expression of genes involved in xenobiotic metabolism and transport. In vitro methods to screen for PXR agonists are used widely. In the current study, computational models for human PXR activators and PXR nonactivators were developed using recursive partitioning (RP), random forest (RF), and support vector machine (SVM) algorithms with VolSurf descriptors. Following 10-fold randomization, the models correctly predicted 82.6-98.9% of activators and 62.0-88.6% of nonactivators. The models were validated using separate test sets. The overall ( n = 15) test set prediction accuracy for PXR activators with RP, RF, and SVM PXR models is 80-93.3%, representing an improvement over models previously reported. All models were tested with a second test set ( n = 145), and the prediction accuracy ranged from 63 to 67% overall. These test set molecules were found to cover the same area in a principal component analysis plot as the training set, suggesting that the predictions were within the applicability domain. The FlexX docking method combined with logistic regression performed poorly in classifying this PXR test set as compared with RP, RF, and SVM but may be useful for qualitative interpretion of interactions within the LBD. From this analysis, VolSurf descriptors and machine learning methods had good classification accuracy and made reliable predictions within the model applicability domain. These methods could be used for high throughput virtual screening to assess for PXR activation, prior to in vitro testing to predict potential drug-drug interactions.