基于化合物结构预测人体ADME/PK性质的效能评价

目的：整合定量结构性质关系（QSPR）模型预测化合物在人体的吸收、分布、代谢、排泄 （ADME）性质参数和基于生理药代动力学（PBPK）模型预测人体药代动力学（PK）曲线的方法,并评价该方法的预测能力。方法：以文献报道的具有体外实测理化、生物药剂学性质和临床观测PK性质的 14个化合物作为模型药物。采用ADMET Predictor软件的QSPR模型预测各个化合物的理化与生物药剂学参数,将上述预测的参数加载到GastroPlus软件的PBPK模型中预测各个化合物经口服给药后在人体的PK 曲线以及主要PK参数。对比预测与实测ADME/PK参数间的差异,以评估所用模型的预测效能。结果： QSPR模型预测的理化与生物药剂学性质参数与观测值间的绝对值较为接近,两者具有较好的线性关系（大部分参数的相关系数均接近或超过0.7）;14个化合物中,有6个化合物（43%）的最大血药浓度 （C_max）预测值落在观测值的2倍误差范围内,9个化合物（64%）的C_max落在观测值的3倍误差范围内; 有7个化合物（50%）的血药浓度-时间曲线下的面积（AUC）预测值落在观测值的2倍误差范围内,8个化合物（57%）的AUC落在观测值的3倍误差范围内。结论：联合QSPR和PBPK模型可用于评估化合物的ADME性质并进一步预测人体PK特征。经过当前工作的验证,表明该方法具有较高的预测能力。     

Comparison of MC4PC and MDL-QSAR rodent carcinogenicity predictions and the enhancement of predictive performance by combining QSAR models

This report presents a comparison of the predictive performance of MC4PC and MDL-QSAR software as well as a method for combining the predictions from both programs to increase overall accuracy. The conclusions are based on 10 x 10% leave-many-out internal cross-validation studies using 1540 training set compounds with 2-year rodent carcinogenicity findings. The models were generated using the same weight of evidence scoring method previously developed [Matthews, E.J., Contrera, J.F., 1998. A new highly specific method for predicting the carcinogenic potential of pharmaceuticals in rodents using enhanced MCASE QSAR-ES software. Regul. Toxicol. Pharmacol. 28, 242-264.]. Although MC4PC and MDL-QSAR use different algorithms, their overall predictive performance was remarkably similar. Respectively, the sensitivity of MC4PC and MDL-QSAR was 61 and 63%, specificity was 71 and 75%, and concordance was 66 and 69%. Coverage for both programs was over 95% and receiver operator characteristic (ROC) intercept statistic values were above 2.00. The software programs had complimentary coverage with none of the 1540 compounds being uncovered by both MC4PC and MDL-QSAR. Merging MC4PC and MDL-QSAR predictions improved the overall predictive performance. Consensus sensitivity increased to 67%, specificity to 84%, concordance to 76%, and ROC to 4.31. Consensus rules can be tuned to reflect the priorities of the user, so that greater emphasis may be placed on predictions with high sensitivity/low false negative rates or high specificity/low false positive rates. Sensitivity was optimized to 75% by reclassifying all compounds predicted to be positive in MC4PC or MDL-QSAR as positive, and specificity was optimized to 89% by reclassifying all compounds predicted negative in MC4PC or MDL-QSAR as negative.     

In vitro to in vivo extrapolation and species response comparisons for drug-induced liver injury (DILI) using DILIsym?: a mechanistic, mathematical model of DILI

Drug-induced liver injury (DILI) is not only a major concern for all patients requiring drug therapy, but also for the pharmaceutical industry. Many new in vitro assays and pre-clinical animal models are being developed to help screen compounds for the potential to cause DILI. This study demonstrates that mechanistic, mathematical modeling offers a method for interpreting and extrapolating results. The DILIsym? model (version 1A), a mathematical representation of DILI, was combined with in vitro data for the model hepatotoxicant methapyrilene (MP) to carry out an in vitro to in vivo extrapolation. In addition, simulations comparing DILI responses across species illustrated how modeling can aid in selecting the most appropriate pre-clinical species for safety testing results relevant to humans. The parameter inputs used to predict DILI for MP were restricted to in vitro inputs solely related to ADME (absorption, distribution, metabolism, elimination) processes. MP toxicity was correctly predicted to occur in rats, but was not apparent in the simulations for humans and mice (consistent with literature). When the hepatotoxicity of MP and acetaminophen (APAP) was compared across rats, mice, and humans at an equivalent dose, the species most susceptible to APAP was not susceptible to MP, and vice versa. Furthermore, consideration of variability in simulated population samples (SimPops?) provided confidence in the predictions and allowed examination of the biological parameters most predictive of outcome. Differences in model sensitivity to the parameters were related to species differences, but the severity of DILI for each drug/species combination was also an important factor.     

Comparison of MC4PC and MDL-QSAR rodent carcinogenicity predictions and the enhancement of predictive performance by combining QSAR models

This report presents a comparison of the predictive performance of MC4PC and MDL-QSAR software as well as a method for combining the predictions from both programs to increase overall accuracy. The conclusions are based on 10 × 10% leave-many-out internal cross-validation studies using 1540 training set compounds with 2-year rodent carcinogenicity findings. The models were generated using the same weight of evidence scoring method previously developed [Matthews, E.J., Contrera, J.F., 1998. A new highly specific method for predicting the carcinogenic potential of pharmaceuticals in rodents using enhanced MCASE QSAR-ES software. Regul. Toxicol. Pharmacol. 28, 242–264.]. Although MC4PC and MDL-QSAR use different algorithms, their overall predictive performance was remarkably similar. Respectively, the sensitivity of MC4PC and MDL-QSAR was 61 and 63%, specificity was 71 and 75%, and concordance was 66 and 69%. Coverage for both programs was over 95% and receiver operator characteristic (ROC) intercept statistic values were above 2.00. The software programs had complimentary coverage with none of the 1540 compounds being uncovered by both MC4PC and MDL-QSAR. Merging MC4PC and MDL-QSAR predictions improved the overall predictive performance. Consensus sensitivity increased to 67%, specificity to 84%, concordance to 76%, and ROC to 4.31. Consensus rules can be tuned to reflect the priorities of the user, so that greater emphasis may be placed on predictions with high sensitivity/low false negative rates or high specificity/low false positive rates. Sensitivity was optimized to 75% by reclassifying all compounds predicted to be positive in MC4PC or MDL-QSAR as positive, and specificity was optimized to 89% by reclassifying all compounds predicted negative in MC4PC or MDL-QSAR as negative.     

Comparative evaluation of 11 in silico models for the prediction of small molecule mutagenicity: role of steric hindrance and electron-withdrawing groups

The goal of this investigation was to perform a comparative analysis on how accurately 11 routinely-used in silico programs correctly predicted the mutagenicity of test compounds that contained either bulky or electron-withdrawing substituents. To our knowledge this is the first study of its kind in the literature. Such substituents are common in many pharmaceutical agents so there is a significant need for reliable in silico programs to predict precisely whether they truly pose a risk for mutagenicity. The predictions from each program were compared to experimental data derived from the Ames II test, a rapid reverse mutagenicity assay with a high degree of agreement with the traditional Ames assay. Eleven in silico programs were evaluated and compared: Derek for Windows, Derek Nexus, Leadscope Model Applier (LSMA), LSMA featuring the in vitro microbial Escherichia coli–Salmonella typhimurium TA102 A-T Suite (LSMA+), TOPKAT, CAESAR, TEST, ChemSilico (±S9 suites), MC4PC and a novel DNA docking model. The presence of bulky or electron-withdrawing functional groups in the vicinity of a mutagenic toxicophore in the test compounds clearly affected the ability of each in silico model to predict non-mutagenicity correctly. This was because of an over reliance on the part of the programs to provide mutagenicity alerts when a particular toxicophore is present irrespective of the structural environment surrounding the toxicophore. From this investigation it can be concluded that these models provide a high degree of specificity (ranging from 71% to 100%) and are generally conservative in their predictions in terms of sensitivity (ranging from 5% t o 78%). These values are in general agreement with most other comparative studies in the literature. Interestingly, the DNA docking model was the most sensitive model evaluated, suggesting a potentially useful new mode of screening for mutagens. Another important finding was that the combination of a quantitative structure–activity relationship and an expert rules system appeared to offer little advantage in terms of sensitivity, despite of the requirement for such a screening paradigm under the ICH M7 regulatory guideline.     

Computer-assisted methods in chemical toxicity prediction

In Silico predictive ADME/Tox screening of compounds is one of the hottest areas in drug discovery. To provide predictions of compound drug-like characteristics early in modern drug-discovery decision making, computational technologies have been widely accepted to develop rapid high throughput in silico ADMET analysis. It is widely perceived that the early screening of chemical entities can significantly reduce the expensive costs associated with late stage failures of drugs due to poor ADME/Tox properties. Drug toxic effects are broadly defined to include toxicity, mutagenicity, carcinogenicity, teratogenicity, neurotoxicity and immunotoxicity. Toxicity prediction techniques and structure-activity relationships relies on the accurate estimation and representation of physico-chemical and toxicological properties. This review highlights some of the freely and commercially available softwares for toxicity predictions. The information content can be utilized as a guide for the scientists involved in the drug discovery arena.     

Experimental versus theoretical log D7.4, pKa and plasma protein binding values for benzodiazepines appearing as new psychoactive substances

The misuse of benzodiazepines as new psychoactive substances is an increasing problem around the world. Basic physicochemical and pharmacokinetic data is required on these substances to interpret and predict their effects upon humans. Experimental log D_7.4, pK_a and plasma protein binding values were determined for 11 benzodiazepines that have recently appeared as new psychoactive substances (3‐hydroxyphenazepam, 4′‐chlorodiazepam, desalkylflurazepam, deschloroetizolam, diclazepam, etizolam, flubromazepam, flubromazolam, meclonazepam, phenazepam, and pyrazolam) and compared with values generated by various software packages (ACD/I‐lab, MarvinSketch, ADMET Predictor and PreADMET). ACD/I‐LAB returned the most accurate values for log D_7.4 and plasma protein binding while ADMET Predictor returned the most accurate values for pK_a. Large variations in predictive errors were observed between compounds. Experimental values are currently preferable and desirable as they may aid with the future ‘training’ of predictive models for these new psychoactive substances.     

In silico prediction of ADMET properties: how far have we come?

There have been considerable advances in the last few years in both the quantity and the quality of in silico ADMET property predictions. Most ADMET properties are now computable, and the accuracy of some of the software predictions for physicochemical properties in particular is close to that of measured data. There is, however, universal agreement that more good experimental ADMET data are needed for use in in silico model development, for models are only as good as the data on which they are based. Many data remain confidential but it is to be hoped that, with projects such as the Vitic toxicity database, being developed by Lhasa Limited, pharmaceutical companies will be prepared to release data to an 'honest broker' on a confidential basis, so that better in silico models can be developed. Incorporation of calculated ADMET properties into drug discovery and development is a multi-factorial problem and really needs a multi-factorial solution. Some progress is being made in this direction and it is hoped that within the foreseeable future software will be available for this purpose.     

An alternative approach for the safety evaluation of new and existing chemicals, an exercise in integrated testing

Various in vitro and in silico methods without animals were applied to 10 substances listed on ELINCS with a complete VIIA base-set available at NOTOX. The hazard assessment for these substances was performed on basis of available non-animal data, QSAR, PBBK-modelling and additional, new in vitro testing was applied. Based on these data predictions on fish toxicity, acute toxicity, skin- and eye-irritation, sensitisation, and toxicity after repeated dosing were made. The predictions were compared with the outcome of the in vivo tests. Nine out of ten predictions on fish LC(50) proved to be correct. For skin- and eye-irritation 70% was predicted correctly. Sensitisation was predicted correctly for 7 out of 10 substances, but three false negatives were found. Acute oral toxicity (LD(50)) and repeated dose toxicity were less successful (5 out of 10 and 2 out of 10 correct predictions, respectively); application of the PBBK model proved successful. Acute dermal toxicity was predicted correctly in 9 out of 10 cases. In general an over-estimation of systemic toxicity was found, which can be explained by an over-prediction of cytotoxicity and worst case assumptions on absorption and binding to (plasma) proteins. This integrated approach leads to a 38% reduction of laboratory animals.     

Prediction of anti-plasmodial activity of Artemisia annua extracts: application of 1H NMR spectroscopy and chemometrics

We describe the application of 1H NMR spectroscopy and chemometrics to the analysis of extracts of Artemisia annua. This approach allowed the discrimination of samples from different sources, and to classify them according to anti-plasmodial activity without prior knowledge of this activity. The use of partial least squares analysis allowed the prediction of actual values for anti-plasmodial activities for independent samples not used in producing the models. The models were constructed using approximately 70% of the samples, with 30% used as a validation set for which predictions were made. Models generally explained >90% of the variance, R(2) in the model, and had a predictive ability, Q(2) of >0.8. This approach was also able to correlate 1H NMR spectra with cytotoxicity (R2=0.9, Q2=0.8). This work demonstrates the potential of NMR spectroscopy and chemometrics for the development of predictive models of anti-plasmodial activity.     

Computational models for cytochrome P450: a predictive electronic model for aromatic oxidation and hydrogen atom abstraction.

Experimental observations suggest that electronic characteristics play a role in the rates of substrate oxidation for cytochrome P450 enzymes. For example, the tendency for oxidation of a certain functional group generally follows the relative stability of the radicals that are formed (e.g., N-dealkylation > O-dealkylation > 2 degrees carbon oxidation > 1 degree carbon oxidation). In addition, results show that useful correlations between the rates of product formation can be developed using electronic models. In this article, we attempt to determine whether a combined computational model for aromatic and aliphatic hydroxylation can be developed. Toward this goal, we used a combination of experimental data and semiempirical molecular orbital calculations to predicted activation energies for aromatic and aliphatic hydroxylation. The resulting model extends the predictive capacity of our previous aliphatic hydroxylation model to include the second most important group of oxidations, aromatic hydroxylation. The combined model can account for about 83% of the variance in the data for the 20 compounds in the training set and has an error of about 0.7 kcal/mol.     

Evaluation of the GastroPlus™ Advanced Compartmental and Transit (ACAT) Model in Early Discovery

Purpose

The aim of this study was to evaluate the oral exposure predictions obtained early in drug discovery with a generic GastroPlus Advanced Compartmental And Transit (ACAT) model based on the in vivo intravenous blood concentration-time profile, in silico properties (lipophilicity, pKa) and in vitro high-throughput absorption-distribution-metabolism-excretion (ADME) data (as determined by PAMPA, solubility, liver microsomal stability assays).

Methods

The model was applied to a total of 623 discovery molecules and their oral exposure was predicted in rats and/or dogs. The predictions of C_max, AUC_last and T_max were compared against the observations.

Results

The generic model proved to make predictions of oral C_max, AUC_last and T_max within 3-fold of the observations for rats in respectively 65%, 68% and 57% of the 537 cases. For dogs, it was respectively 77%, 79% and 85% of the 124 cases. Statistically, the model was most successful at predicting oral exposure of Biopharmaceutical Classification System (BCS) class 1 compounds compared to classes 2 and 3, and was worst at predicting class 4 compounds oral exposure.

Conclusion

The generic GastroPlus ACAT model provided reasonable predictions especially for BCS class 1 compounds. For compounds of other classes, the model may be refined by obtaining more information on solubility and permeability in secondary assays. This increases confidence that such a model can be used in discovery projects to understand the parameters limiting absorption and extrapolate predictions across species. Also, when predictions disagree with the observations, the model can be updated to test hypotheses and understand oral absorption.

     

Exploiting QSAR models in lead optimization

QSAR models can play a vital role in both the opening phase and the endgame of lead optimization. In the opening phase, there is often a large quantity of data from high-throughput screening (HTS), and potential leads need to be selected from several distinct chemotypes. In the endgame, the throughput of the final, critical ADMET and pharmacokinetic assays is often not sufficient to allow full experimental characterization of all the structures in the available time. A considerable amount of the current research toward new QSAR models is based on the modeling of the general ADMET phenomena, with the aim of constructing globally applicable models. The process to construct QSAR models is relatively straightforward; however, it is also simple to build misleading, or even incorrect, models. This review considers the key developments in the field of QSAR modeling: how QSAR models are constructed, how they can be validated, their reliability and their applicability. If applied carefully and appropriately, the QSAR technique has a valuable role to play during lead optimization.     

A computationally based identification algorithm for estrogen receptor ligands: part 2. Evaluation of a hERalpha binding affinity model.

The objective of this study was to evaluate the capability of an expert system described in the previous paper (S. Bradbury et al., Toxicol. Sci. 58, 253-269) to identify the potential for chemicals to act as ligands of mammalian estrogen receptors (ERs). The basis of the expert system was a structure activity relationship (SAR) model, based on relative binding affinity (RBA) values for steroidal and nonsteroidal chemicals derived from human ERalpha (hERalpha) competitive binding assays. The expert system enables categorization of chemicals into (RBA ranges of < 0.1, 0.1 to 1, 1 to 10, 10 to 100, and >150% relative to 17ss-estradiol. In the current analysis, the algorithm was evaluated with respect to predicting RBAs of chemicals assayed with ERs from MCF7 cells, and mouse and rat uterine preparations. The best correspondence between predicted and observed RBA ranges was obtained with MCF7 cells. The agreement between predictions from the expert system and data from binding assays with mouse and rat ER(s) were less reliable, especially for chemicals with RBAs less than 10%. Prediction errors often were false positives, i.e., predictions of greater than observed RBA values. While discrepancies were likely due, in part, to species-specific variations in ER structure and ligand binding affinity, a systematic bias in structural characteristics of chemicals in the hERalpha training set, compared to the rodent evaluation data sets, also contributed to prediction errors. False-positive predictions were typically associated with ligands that had shielded electronegative sites. Ligands with these structural characteristics were not well represented in the training set used to derive the expert system. Inclusion of a shielding criterion into the original expert system significantly increased the accuracy of RBA predictions. With this additional structural requirement, 38 of 46 compounds with measured RBA values greater than 10% in hERalpha, MCF7, and rodent uterine preparations were correctly categorized. Of the remaining 129 compounds in the combined data sets, RBA values for 65 compounds were correctly predicted, with 47 of the incorrect predictions being false positives. Based upon this exploratory analysis, the modeling approach, combined with a high-quality training set of RBA values derived from a diverse set of chemical structures, could provide a credible tool for prioritizing chemicals with moderate to high ER binding affinity for subsequent in vitro or in vivo assessments.     

Prediction of short-term antidepressant response using probabilistic graphical models with replication across multiple drugs and treatment settings

Heterogeneity in the clinical presentation of major depressive disorder and response to antidepressants limits clinicians’ ability to accurately predict a specific patient’s eventual response to therapy. Validated depressive symptom profiles may be an important tool for identifying poor outcomes early in the course of treatment. To derive these symptom profiles, we first examined data from 947 depressed subjects treated with selective serotonin reuptake inhibitors (SSRIs) to delineate the heterogeneity of antidepressant response using probabilistic graphical models (PGMs). We then used unsupervised machine learning to identify specific depressive symptoms and thresholds of improvement that were predictive of antidepressant response by 4 weeks for a patient to achieve remission, response, or nonresponse by 8 weeks. Four depressive symptoms (depressed mood, guilt feelings and delusion, work and activities and psychic anxiety) and specific thresholds of change in each at 4 weeks predicted eventual outcome at 8 weeks to SSRI therapy with an average accuracy of 77% (p = 5.5E-08). The same four symptoms and prognostic thresholds derived from patients treated with SSRIs correctly predicted outcomes in 72% (p = 1.25E-05) of 1996 patients treated with other antidepressants in both inpatient and outpatient settings in independent publicly-available datasets. These predictive accuracies were higher than the accuracy of 53% for predicting SSRI response achieved using approaches that (i) incorporated only baseline clinical and sociodemographic factors, or (ii) used 4-week nonresponse status to predict likely outcomes at 8 weeks. The present findings suggest that PGMs providing interpretable predictions have the potential to enhance clinical treatment of depression and reduce the time burden associated with trials of ineffective antidepressants. Prospective trials examining this approach are forthcoming.Subject terms: Translational research, Depression     

Evaluation of drug–drug interactions for oncology therapies: in vitro–in vivo extrapolation model-based risk assessment

Aims

Understanding drug–drug interactions (DDI) is a critical part of the drug development process as polypharmacy has become commonplace in many therapeutic areas including the cancer patient population. The objectives of this study were to investigate cytochrome P450 (CYP)-mediated DDI profiles available for therapies used in the oncology setting and evaluate how models based on in vitro–in vivo extrapolation performed in predicting CYP-mediated DDI risk.

Methods

A dataset of 125 oncology therapies was collated using drug label and approval history information, incorporating in vitro and clinical PK data. The predictive accuracy of the basic and net effect mechanistic static models was assessed using this oncology drug dataset, for both victim and perpetrator potential of CYP3A-mediated DDI.

Results

The incidence of CYP3A-mediated interaction potential was 47%, 22% and 11% for substrates, inhibitors and inducers, respectively. The basic models for precipitants gave conservative predictions with no false negatives, whilst the mechanistic static models provided reasonable quantitative predictions (2.3–3-fold error). Further analysis revealed that incorporating DDI at the level of the intestine was in most cases over-predicting interaction magnitude due to overestimates of the rate and extent of oral absorption of the precipitant. Quantifying victim DDI potential was also demonstrated using f_mCYP3A estimates from ketoconazole clinical DDI studies to predict the magnitude of interaction on co-administration with the CYP3A inducer, rifampicin (1.6–3.3 fold error).

Conclusions

This work illustrates the utility and limitations of current DDI risk assessment approaches applied to a range of contemporary anti-cancer agents, and discusses the implications for therapeutic combination strategies.