Mixed learning algorithms and features ensemble in hepatotoxicity prediction

Drug-induced liver injury, although infrequent, is an important safety concern that can lead to fatality in patients and failure in drug developments. In this study, we have used an ensemble of mixed learning algorithms and mixed features for the development of a model to predict hepatic effects. This robust method is based on the premise that no single learning algorithm is optimum for all modelling problems. An ensemble model of 617 base classifiers was built from a diverse set of 1,087 compounds. The ensemble model was validated internally with five-fold cross-validation and 25 rounds of y-randomization. In the external validation of 120 compounds, the ensemble model had achieved an accuracy of 75.0%, sensitivity of 81.9% and specificity of 64.6%. The model was also able to identify 22 of 23 withdrawn drugs or drugs with black box warning against hepatotoxicity. Dronedarone which is associated with severe liver injuries, announced in a recent FDA drug safety communication, was predicted as hepatotoxic by the ensemble model. It was found that the ensemble model was capable of classifying positive compounds (with hepatic effects) well, but less so on negatives compounds when they were structurally similar. The ensemble model built in this study is made available for public use.     

Quantifying the hydrophobic effect. 1. A computer simulation-molecular-thermodynamic model for the self-assembly of hydrophobic and amphiphilic solutes in aqueous solution

Surfactant micellization and micellar solubilization in aqueous solution can be modeled using a molecular-thermodynamic (MT) theoretical approach; however, the implementation of MT theory requires an accurate identification of the portions of solutes (surfactants and solubilizates) that are hydrated and unhydrated in the micellar state. For simple solutes, such identification is comparatively straightforward using simple rules of thumb or group-contribution methods, but for more complex solutes, the hydration states in the micellar environment are unclear. Recently, a hybrid method was reported by these authors in which hydrated and unhydrated states are identified by atomistic simulation, with the resulting information being used to make MT predictions of micellization and micellar solubilization behavior. Although this hybrid method improves the accuracy of the MT approach for complex solutes with a minimum of computational expense, the limitation remains that individual atoms are modeled as being in only one of two states-head or tail-whereas in reality, there is a continuous spectrum of hydration states between these two limits. In the case of hydrophobic or amphiphilic solutes possessing more complex chemical structures, a new modeling approach is needed to (i) obtain quantitative information about changes in hydration that occur upon aggregate formation, (ii) quantify the hydrophobic driving force for self-assembly, and (iii) make predictions of micellization and micellar solubilization behavior. This article is the first in a series of articles introducing a new computer simulation-molecular thermodynamic (CS-MT) model that accomplishes objectives (i)-(iii) and enables prediction of micellization and micellar solubilization behaviors, which are infeasible to model directly using atomistic simulation. In this article (article 1 of the series), the CS-MT model is introduced and implemented to model simple oil aggregates of various shapes and sizes, and its predictions are compared to those of the traditional MT model. The CS-MT model is formulated to allow the prediction of the free-energy change associated with aggregate formation (gform) of solute aggregates of any shape and size by performing only two computer simulations-one of the solute in bulk water and the other of the solute in an aggregate of arbitrary shape and size. For the 15 oil systems modeled in this article, the average discrepancy between the predictions of the CS-MT model and those of the traditional MT model for gform is only 1.04%. In article 2, the CS-MT modeling approach is implemented to predict the micellization behavior of nonionic surfactants; in article 3, it is used to predict the micellization behavior of ionic and zwitterionic surfactants.     

Cross-validation of component models: A critical look at current methods

In regression, cross-validation is an effective and popular approach that is used to decide, for example, the number of underlying features, and to estimate the average prediction error. The basic principle of cross-validation is to leave out part of the data, build a model, and then predict the left-out samples. While such an approach can also be envisioned for component models such as principal component analysis (PCA), most current implementations do not comply with the essential requirement that the predictions should be independent of the entity being predicted. Further, these methods have not been properly reviewed in the literature. In this paper, we review the most commonly used generic PCA cross-validation schemes and assess how well they work in various scenarios.     

Theoretical Calculations and Hydrothermal Processing of BaWO<Subscript>4</Subscript> Materials Under Environmentally Friendly Conditions

The hydrothermal processing conditions of BaWO₄ was studied by using a thermodynamic model of electrolytic solutions in order to avoid the empirical trial-and-error mode for optimizing synthesis of this material. The approach used makes it possible to predict the optimum conditions through stability and yield diagrams, which relate the equilibrium concentration of all species present as a function of temperature, solution pH, and input reagent concentrations. The theoretical predictions were verified by experiments at the predicted optimum conditions. The results showed that the thermodynamic model is adequate to predict the hydrothermal synthesis conditions under which BaWO₄ is stable and can be obtained as a phase-pure form.     

Hazard Evaluation Support System (HESS) for predicting repeated dose toxicity using toxicological categories

Repeated dose toxicity (RDT) is one of the most important hazard endpoints in the risk assessment of chemicals. However, due to the complexity of the endpoints associated with whole body assessment, it is difficult to build up a mechanistically transparent structure–activity model. The category approach, based on mechanism information, is considered to be an effective approach for data gap filling for RDT by read-across. Therefore, a library of toxicological categories was developed using experimental RDT data for 500 chemicals and mechanistic knowledge of the effects of these chemicals on different organs. As a result, 33 categories were defined for 14 types of toxicity, such as hepatotoxicity, hemolytic anemia, etc. This category library was then incorporated in the Hazard Evaluation Support System (HESS) integrated computational platform to provide mechanistically reasonable predictions of RDT values for untested chemicals. This article describes the establishment of a category library and the associated HESS functions used to facilitate the mechanistically reasonable grouping of chemicals and their subsequent read-across.     

Combined effect of temperature and organic modifier concentration on the retention under single mode gradient conditions in reversed-phase HPLC

The combined effect of temperature, T, and organic modifier concentration, phi, on the retention under gradient conditions in RPLC is studied by considering, both theoretically and experimentally gradients, of phi at constant T and gradients of T at constant phi. Two approaches are examined: in the first approach the prediction of the elution time of a sample solute is based on the isocratic/isothermal properties of this solute. The second approach is based on a direct fitting procedure of a proper retention model to 2-D isocratic/T-gradient or isothermal/phi-gradient retention data. These approaches were tested using alkylbenzes in eluting systems modified by ACN. We found that both approaches can give excellent predictions under certain prerequisites. However, the first approach exhibits the notable advantage that it can be used effectively to predict retention times under any kind of phi-gradients at constant T or T-gradients at constant phi. The second approach has the advantage that it is relatively simple but its applicability is very restricted since its predictions are satisfactory only if the gradients are of the same kind with those used in the fitting procedure and the conditions lie within those used for fitting.     

Simulation of the oxidative metabolism of diclofenac by electrochemistry/(liquid chromatography/)mass spectrometry

Diclofenac is a frequently prescribed drug for rheumatic diseases and muscle pain. In rare cases, it may be associated with a severe hepatotoxicity. In literature, it is discussed whether this toxicity is related to the oxidative phase I metabolism, resulting in electrophilic quinone imines, which can subsequently react with nucleophiles present in the liver in form of glutathione or proteins. In this work, electrochemistry coupled to mass spectrometry is used as a tool for the simulation of the oxidative pathway of diclofenac. Using this purely instrumental approach, diclofenac was oxidized in a thin layer cell equipped with a boron doped diamond working electrode. Sum formulae of generated oxidation products were calculated based on accurate mass measurements with deviations below 2 ppm. Quinone imines from diclofenac were detected using this approach. It could be shown for the first time that these quinone imines do not react with glutathione exclusively but also with larger molecules such as the model protein β-lactoglobulin A. A tryptic digest of the generated drug–protein adduct confirms that the protein is modified at the only free thiol-containing peptide. This simple and purely instrumental set-up offers the possibility of generating reactive metabolites of diclofenac and to assess their reactivity rapidly and easily.     

Unified model to predict self-assembly of nonionic surfactants in solution and adsorption on solid or fluid hydrophobic surfaces: effect of molecular structure

We have developed a pseudo-phase model to predict the self-assembly of nonionic surfactants on hydrophobic solid or fluid interfaces and in bulk solution. The uniqueness of this model is that it provides the relationship between molecular structure and self-assembly in solution and on interfaces. This model requires the input of minimal new experimental data. The remaining model parameters may be calculated on the basis of the surfactant molecular structure. The validity of the model has been established by comparing predictions with a wide array of experimental data for nonionic surfactant adsorption at the hydrophobic solid-water interface and at the air-water interface. The same model is then used to predict the self-assembly in bulk solution. The model predictions for critical aggregation concentration, aggregate shapes, and adsorption isotherms of various surfactants are in good agreement with the experimental data available in the literature.     

In vitro metabolites characterization of ponatinib in human liver microsomes using ultra-high performance liquid chromatography combined with Q-Exactive-Orbitrap tandem mass spectrometry

Ponatinib is an oral drug for the treatment of chronic myeloid leukemia and acute lymphoblastic leukemia, which has been reported to increase the risk of hepatotoxicity. The aim of this study was to characterize the metabolites of ponatinib in human liver microsomes as well as its reactive metabolites. Ponatinib was incubated with human liver microsomes in the presence of NADPH and trapping agents (glutathione or potassium cyanide). The metabolites were characterized by liquid chromatography in combination with Q-Exactive-Orbitrap-MS. Under the current conditions, six metabolites were detected and structurally identified on the basis of their accurate masses, fragmentation patterns, and retention times. M3 (N-demethylation) was unambiguously identified by matching its retention time and fragment ions with those of its reference standard. N-demethylation and oxygenation were proved to be the predominant metabolic pathways of ponatinib. In addition, two reactive metabolites (cyano adducts) were detected in human liver microsomes in the presence of potassium cyanide and NADPH, suggesting that ponatinib underwent CYP450-mediated metabolic activation, which could be one of the causative mechanisms for its hepatotoxicity. The current study provides new information regarding the metabolic profiles of ponatinib and would be helpful in understanding the effectiveness and toxicity of ponatinib, especially the mechanism of hepatotoxicity.     

Real‐Time In Vivo Hepatotoxicity Monitoring through Chromophore‐Conjugated Photon‐Upconverting Nanoprobes

Drug toxicity is a long‐standing concern of modern medicine. A typical anti‐pain/fever drug paracetamol often causes hepatotoxicity due to peroxynitrite ONOO⁻. Conventional blood tests fail to offer real‐time unambiguous visualization of such hepatotoxicity in vivo. Here we report a luminescent approach to evaluate acute hepatotoxicity in vivo by chromophore‐conjugated upconversion nanoparticles. Upon injection, these nanoprobes mainly accumulate in the liver and the luminescence of nanoparticles remains suppressed owing to energy transfer to the chromophore. ONOO⁻ can readily bleach the chromophore and thus recover the luminescence, the presence of ONOO⁻ in the liver leads to fast restoring of the near‐infrared emission. Taking advantages of the high tissue‐penetration capability of near‐infrared excitation/emission, these nanoprobes achieve real‐time monitoring of hepatotoxicity in living animals, thereby providing a convenient screening strategy for assessing hepatotoxicity of synthetic drugs.     

Prediction of UV and ESI-MS signal intensities

All major pharmaceutical companies maintain large collections of compounds that are used either for screening against biological targets or as synthetic precursors. The quality assessment of these compounds is typically done by liquid chromatography combined with mass spectroscopy (LC/MS) and UV purity control. To facilitate the analysis of the analytical data, we have built computational models to predict UV and MS signal intensities under experimental LC/MS conditions. The discriminant partial-least-squares technique was used for classifying compounds into those most likely to yield a MS signal and others where the signal is below the detection limit (94% and 88% correct predictions, respectively). In the case of UV prediction, we compared this statistical linear-regression technique to a knowledge-based approach. A combination of both techniques proved to be the most reliable (96/98% correct predictions of UV-active/ UV-inactive compounds). Both models have been incorporated into the automated compound integrity profiling at F. Hoffmann-La Roche.     

Improving confidence in (Q)SAR predictions under Canada’s Chemicals Management Plan – a chemical space approach$

Abstract

One of the key challenges of Canada’s Chemicals Management Plan (CMP) is assessing chemicals with limited/no empirical hazard data for their risk to human health. In some instances, these chemicals have not been tested broadly for their toxicological potency; as such, limited information exists on their potential to induce human health effects following exposure. Although (quantitative) structure activity relationship ((Q)SAR) models are able to generate predictions to address data gaps for certain toxicological endpoints, the confidence in predictions also needs to be addressed. One way to address this issue is to apply a chemical space approach. This approach uses international toxicological databases, for example, those available in the Organisation for Economic Co-operation and Development (OECD) QSAR Toolbox. The approach,assesses a model’s ability to predict the potential hazards of chemicals that have limited hazard data that require assessment under the CMP when compared to a larger, data-rich chemical space that is structurally similar to chemicals of interest. This evaluation of a model’s predictive ability makes (Q)SAR analysis more transparent and increases confidence in the application of these predictions in a risk-assessment context. Using this approach, predictions for such chemicals obtained from four (Q)SAR models were successfully classified into high, medium and low confidence levels to better inform their use in decision-making.     

Estimation of interfacial behavior using the global phase diagram approach: I. Carbon dioxide-n-alkanes

This study looks for the first time at the possibility of predicting the interfacial tension in mixtures without preliminary resource to their experimental data. For this purpose the quantitative global phase diagram (klGPD)-based approach (GPDA), which needs only two or three key experimental points of one homologue for predicting the complete phase behavior in whole homologues series of binary systems, is combined with the gradient theory (GT) methodology. The resulting model is able to predict the data in satisfactory manner, although the increasing asymmetry between the compounds of the mixture probably affects the ability of GPDA to yield accurate predictions of phase equilibria and interface tension simultaneously.     

Numerical and analytical studies of the electrical conductivity of a concentrated colloidal suspension

In the past few years, different models and analytical approximations have been developed facing the problem of the electrical conductivity of a concentrated colloidal suspension, according to the cell-model concept. Most of them make use of the Kuwabara cell model to account for hydrodynamic particle-particle interactions, but they differ in the choice of electrostatic boundary conditions at the outer surface of the cell. Most analytical and numerical studies have been developed using two different sets of boundary conditions of the Neumann or Dirichlet type for the electrical potential, ionic concentrations or electrochemical potentials at that outer surface. In this contribution, we study and compare numerical conductivity predictions with results obtained using different analytical formulas valid for arbitrary zeta potentials and thin double layers for each of the two common sets of boundary conditions referred to above. The conductivity will be analyzed as a function of particle volume fraction, phi, zeta potential, zeta, and electrokinetic radius, kappaa (kappa(-1) is the double layer thickness, and a is the radius of the particle). A comparison with some experimental conductivity results in the literature is also given. We demonstrate in this work that the two analytical conductivity formulas, which are mainly based on Neumann- and Dirichlet-type boundary conditions for the electrochemical potential, predict values of the conductivity very close to their corresponding numerical results for the same boundary conditions, whatever the suspension or solution parameters, under the assumption of thin double layers where these approximations are valid. Furthermore, both analytical conductivity equations fulfill the Maxwell limit for uncharged nonconductive spheres, which coincides with the limit kappaa --> infinity. However, some experimental data will show that the Neumann, either numerical or analytical, approach is unable to make predictions in agreement with experiments, unlike the Dirichlet approach which correctly predicts the experimental conductivity results. In consequence, a deeper study has been performed with numerical and analytical predictions based on Dirichlet-type boundary conditions.     

Hepatoprotective and hepatotoxic actions of oleanolic acid-type triterpenoidal glucuronides on rat primary hepatocyte cultures

The protective effects of oleanolic acid-type saponins and their derivatives on in vitro immunological liver injury of primary cultured rat hepatocytes were studied. A known antihepatotoxic saponin (chikusetsusaponin IVa, 1) showed hepatoprotective activity in this model. Although a rhamnosyl derivative (2) of 1 similarly showed hepatoprotective activity, its prosapogenin (5) did not show any hepatoprotective activity. On the contrary, 5 exhibited cytotoxicity toward liver cells. In the absence of antiserum, monodesmosyl saponins showed hepatotoxicity, while the bisdesmosyl saponins except for 1, did not show such hepatotoxicity. In order to clarify the effects of the sugar residues at C-3 and C-28 responsible for hepatoprotective and hepatotoxic actions, oleanolic acid 3-O-glucuronide (2a) and oleanolic acid 28-O-glucoside (2b) were prepared and tested. 2b showed neither hepatoprotective action nor hepatotoxicity. In contrast, 2a was effective at 90 microM on hepatoprotection, although it showed strong hepatotoxicity. Oleanolic acid (2c) itself showed both hepatoprotective action and weak hepatotoxicity. Therefore, the hepatoprotective activity of these types of saponins could represent a balance between hepatoprotective action and hepatotoxicity.     

Prediction of polymerization shrinkage using molecular modeling

Group interaction modeling (GIM) is used to predict the changes in the volume and, therefore, shrinkage of a wide variety of commonly encountered acrylate and methacrylate monomers during polymerization. The predictions of the model are in excellent agreement with experimental data available in the literature. It is demonstrated that, given appropriate estimations of the ultimate matrix morphology, GIM can be used to give estimates of polymerization shrinkage in three‐dimensional crosslinked polymer networks that relate well to experimental data available for dental composite matrices. GIM provides a very useful framework within which the phenomenon of polymerization shrinkage can be considered. The approach challenges certain common misconceptions relating conversion to shrinkage. The limitations of this approach are discussed, and targets for future research are clearly outlined that could extend the scope of this methodology to encompass novel systems for which morphological information is not available. © 2003 Wiley Periodicals, Inc. J Polym Sci Part B: Polym Phys 41: 528–548, 2003     

Metabolic alterations in triptolide‐induced acute hepatotoxicity

Triptolide, a major active constitute of Tripterygium wilfordii Hook. F, is prescribed for the treatment of autoimmune diseases in China. One of its most severe adverse effects observed in the clinical use is hepatotoxicity, but the mechanism is still unknown. Therefore, the present study applied an LC/MS‐based metabolomic analysis to characterize the metabolomic changes in serum and liver induced by triptolide in mice. Mice were administered triptolide by gavage to establish the acute liver injury model, and serum biochemical and liver histological analyses were applied to assess the degree of toxicity. Multivariate data analyses were performed to investigate the metabolic alterations. Potential metabolites were identified using variable importance in the projection values and Student's t‐test. A total of 30 metabolites were observed that were significantly changed by triptolide treatment and the abundance of 29 metabolites was correlated with the severity of toxicity. Pathway analysis indicated that the mechanism of triptolide‐induced hepatotoxicity was related to alterations in multiple metabolic pathways, including glutathione metabolism, tricarboxylic acid cycle, purine metabolism, glycerophospholipid metabolism, taurine and hypotaurine metabolism, pantothenate and CoA biosynthesis, pyrimidine metabolism and amino acid metabolism. The current study provides new mechanistic insights into the metabolic alterations that lead to triptolide‐induced hepatotoxicity.     

Prediction of protein folding rates from primary sequences using hybrid sequence representation

The ability to predict protein folding rates constitutes an important step in understanding the overall folding mechanisms. Although many of the prediction methods are structure based, successful predictions can also be obtained from the sequence. We developed a novel method called prediction of protein folding rates (PPFR), for the prediction of protein folding rates from protein sequences. PPFR implements a linear regression model for each of the mainstream folding dynamics including two-, multi-, and mixed-state proteins. The proposed method provides predictions characterized by strong correlations with the experimental folding rates, which equal 0.87 for the two- and multistate proteins and 0.82 for the mixed-state proteins, when evaluated with out-of-sample jackknife test. Based on in-sample and out-of-sample tests, the PPFR's predictions are shown to be better than most of other sequence only and structure-based predictors and complementary to the predictions of the most recent sequence-based QRSM method. We show that simultaneous incorporation of several characteristics, including the sequence, physiochemical properties of residues, and predicted secondary structure provides improved quality. This hybridized prediction model was analyzed to reveal the complementary factors that can be used in tandem to predict folding rates. We show that bigger proteins require more time for folding, higher helical and coil content and the presence of Phe, Asn, and Gln may accelerate the folding process, the inclusion of Ile, Val, Thr, and Ser may slow down the folding process, and for the two-state proteins increased beta-strand content may decelerate the folding process. Finally, PPFR provides strong correlation when predicting sequences with low similarity.