Integrating Drug's Mode of Action into Quantitative Structure-Activity Relationships for Improved Prediction of Drug-Induced Liver Injury

Machine Learning to Predict Drug-Induced Liver Injury and Its Validation on Failed Drug Candidates in Development

Drug-induced liver injury (DILI) poses a significant challenge for the pharmaceutical industry and regulatory bodies. Despite extensive toxicological research aimed at mitigating DILI risk, the effectiveness of these techniques in predicting DILI in humans remains limited. Consequently, researchers have explored novel approaches and procedures to enhance the accuracy of DILI risk prediction for drug candidates under development. In this study, we leveraged a large human dataset to develop machine learning models for assessing DILI risk. The performance of these prediction models was rigorously evaluated using a 10-fold cross-validation approach and an external test set. Notably, the random forest (RF) and multilayer perceptron (MLP) models emerged as the most effective in predicting DILI. During cross-validation, RF achieved an average prediction accuracy of 0.631, while MLP achieved the highest Matthews Correlation Coefficient (MCC) of 0.245. To validate the models externally, we applied them to a set of drug candidates that had failed in clinical development due to hepatotoxicity. Both RF and MLP accurately predicted the toxic drug candidates in this external validation. Our findings suggest that in silico machine learning approaches hold promise for identifying DILI liabilities associated with drug candidates during development.

Deep Learning Algorithm Based on Molecular Fingerprint for Prediction of Drug-Induced Liver Injury

Drug-induced liver injury (DILI) is one the rare adverse drug reaction (ADR) and multifactorial endpoints. Current preclinical animal models struggle to anticipate it, and in silico methods have emerged as a way with significant potential for doing so. In this study, a high-quality dataset of 1573 compounds was assembled. The 48 classification models, which depended on six different molecular fingerprints, were built via deep neural network (DNN) and seven machine learning algorithms. Comparing the results of the DNN and machine learning models, the optional performing model was found as the one developed based on the DNN with ECFP_6 as input, which achieved the area under the receiver operating characteristic curve (AUC) of 0.713, balanced accuracy (BA) of 0.680, and F1 of 0.753. In addition, we used the SHapley Additive exPlanations (SHAP) algorithm to interpret the models, identified the crucial structural fragments related to DILI risk, and selected the top ten substructures with the highest contribution rankings to serve as warning indicators for subsequent drug hepatotoxicity screening studies. The study demonstrates that the DNN models developed based on molecular fingerprints can be a trustworthy and efficient tool for determining the risk of DILI during the pre-development of novel medications.

Computational models for predicting liver toxicity in the deep learning era

Drug-induced liver injury (DILI) is a severe adverse reaction caused by drugs and may result in acute liver failure and even death. Many efforts have centered on mitigating risks associated with potential DILI in humans. Among these, quantitative structure-activity relationship (QSAR) was proven to be a valuable tool for early-stage hepatotoxicity screening. Its advantages include no requirement for physical substances and rapid delivery of results. Deep learning (DL) made rapid advancements recently and has been used for developing QSAR models. This review discusses the use of DL in predicting DILI, focusing on the development of QSAR models employing extensive chemical structure datasets alongside their corresponding DILI outcomes. We undertake a comprehensive evaluation of various DL methods, comparing with those of traditional machine learning (ML) approaches, and explore the strengths and limitations of DL techniques regarding their interpretability, scalability, and generalization. Overall, our review underscores the potential of DL methodologies to enhance DILI prediction and provides insights into future avenues for developing predictive models to mitigate DILI risk in humans.

Identifying chemicals based on receptor binding/bioactivation/mechanistic explanation associated with potential to elicit hepatotoxicity and to support structure activity relationship-based read-across

The liver is the most common target organ in toxicology studies. The development of chemical structural alerts for identifying hepatotoxicity will play an important role in in silico model prediction and help strengthen the identification of analogs used in structure activity relationship (SAR)- based read-across. The aim of the current study is development of an SAR-based expert-system decision tree for screening of hepatotoxicants across a wide range of chemistry space and proposed modes of action for clustering of chemicals using defined core chemical categories based on receptor-binding or bioactivation. The decision tree is based on ∼ 1180 different chemicals that were reviewed for hepatotoxicity information. Knowledge of chemical receptor binding, metabolism and mechanistic information were used to group these chemicals into 16 different categories and 102 subcategories: four categories describe binders to 9 different receptors, 11 categories are associated with possible reactive metabolites (RMs) and there is one miscellaneous category. Each chemical subcategory has been associated with possible modes of action (MOAs) or similar key structural features. This decision tree can help to screen potential liver toxicants associated with core structural alerts of receptor binding and/or RMs and be used as a component of weight of evidence decisions based on SAR read-across, and to fill data gaps.

Estimating drug-induced liver injury risk by in vitro molecular initiation response and pharmacokinetic parameters for during early drug development

Drug-induced liver injury (DILI) is a major factor influencing new drug withdrawal; therefore, an appropriate toxicity assessment at the preclinical stage is required. Previous in silico models have been established using compound information listed in large data sources, thereby limiting the DILI risk prediction for new drugs. Herein, we first constructed a model to predict DILI risk based on a molecular initiating event (MIE) predicted by quantitative structure-activity relationships, admetSAR parameters (e.g. cytochrome P450 reactivity, plasma protein binding, and water-solubility), and clinical information (maximum daily dose [MDD] and reactive metabolite [RM]) for 186 compounds. The accuracy of the models using MIE, MDD, RM, and admetSAR alone were 43.2%, 47.3%, 77.0%, and 68.9%, while the "predicted MIE + admetSAR + MDD + RM" model's accuracy was 75.7%. The contribution of MIE to the overall prediction accuracy was little effect or rather worsening it. However, it was considered that MIE was a valuable parameter and that it contributed to detect high DILI risk compounds in the early development stage. We next examined the effect of stepwise changes in MDD on altering the DILI risk and estimating the maximum safety dose (MSD) for clinical use based on structural information, admetSAR, and MIE parameters because it is important to estimate the dose that could prevent the DILI onset in clinical conditions. Low-MSD compounds might increase the DILI risk, as these compounds were classified as "most-DILI concern" at low doses. In conclusion, MIE parameters were especially useful to check the DILI concern compounds and to prevent the underestimation of DILI risk in the early stage of drug development.

Supervised chemical graph mining improves drug-induced liver injury prediction

Drug-induced liver injury (DILI) is the main cause of drug failure in clinical trials. The characterization of toxic compounds in terms of chemical structure is important because compounds can be metabolized to toxic substances in the liver. Traditional machine learning approaches have had limited success in predicting DILI, and emerging deep graph neural network (GNN) models are yet powerful enough to predict DILI. In this study, we developed a completely different approach, supervised subgraph mining (SSM), a strategy to mine explicit subgraph features by iteratively updating individual graph transitions to maximize DILI fidelity. Our method outperformed previous methods including state-of-the-art GNN tools in classifying DILI on two different datasets: DILIst and TDC-benchmark. We also combined the subgraph features by using SMARTS-based frequent structural pattern matching and associated them with drugs' ATC code.

Applications of In Silico Models to Predict Drug-Induced Liver Injury

Drug-induced liver injury (DILI) is a major cause of the withdrawal of pre-marketed drugs, typically attributed to oxidative stress, mitochondrial damage, disrupted bile acid homeostasis, and innate immune-related inflammation. DILI can be divided into intrinsic and idiosyncratic DILI with cholestatic liver injury as an important manifestation. The diagnosis of DILI remains a challenge today and relies on clinical judgment and knowledge of the insulting agent. Early prediction of hepatotoxicity is an important but still unfulfilled component of drug development. In response, in silico modeling has shown good potential to fill the missing puzzle. Computer algorithms, with machine learning and artificial intelligence as a representative, can be established to initiate a reaction on the given condition to predict DILI. DILIsym is a mechanistic approach that integrates physiologically based pharmacokinetic modeling with the mechanisms of hepatoxicity and has gained increasing popularity for DILI prediction. This article reviews existing in silico approaches utilized to predict DILI risks in clinical medication and provides an overview of the underlying principles and related practical applications.

Relationships between the Structure and Severe Drug-Induced Liver Injury for Low, Medium, and High Doses of Drugs

Assessment of structure-activity relationships (SARs) for predicting severe drug-induced liver injury (DILI) is essential since in vivo and in vitro preclinical methods cannot detect many druglike compounds disrupting liver functions. To date, plenty of SAR models for the prediction of DILI have been developed; however, none of them considered the route of drug administration and daily dose, which may introduce significant bias into prediction results. We have created a dataset of 617 drugs with parenteral and oral administration routes and consistent information on DILI severity. We have found a clear relationship between route, dose, and DILI severity. According to SAR, nearly 40% of moderate- and non-DILI-causing drugs would cause severe DILI if they were administered at high oral doses. We have proposed the following approach to predict severe DILI. New compounds recommended to be used at low oral doses (<∼10 mg daily), or parenterally, can be considered not causing severe DILI. DILI for compounds administered at medium oral doses (∼10-100 mg daily; 22.2% of drugs under consideration) can be considered unpredictable because reasonable SAR models were not obtained due to the small size and heterogeneity of the corresponding dataset. The DILI potential of the compounds recommended to be used at high oral doses (more than ∼100 mg daily) can be estimated using SAR modeling. The balanced accuracy of the approach calculated by a 10-fold cross-validation procedure is 0.803. The developed approach can be used to estimate severe DILI for druglike compounds proposed to use at low and high oral doses or parenterally at the early stages of drug development.

In silico approaches in organ toxicity hazard assessment: current status and future needs in predicting liver toxicity

Hepatotoxicity is one of the most frequently observed adverse effects resulting from exposure to a xenobiotic. For example, in pharmaceutical research and development it is one of the major reasons for drug withdrawals, clinical failures, and discontinuation of drug candidates. The development of faster and cheaper methods to assess hepatotoxicity that are both more sustainable and more informative is critically needed. The biological mechanisms and processes underpinning hepatotoxicity are summarized and experimental approaches to support the prediction of hepatotoxicity are described, including toxicokinetic considerations. The paper describes the increasingly important role of in silico approaches and highlights challenges to the adoption of these methods including the lack of a commonly agreed upon protocol for performing such an assessment and the need for in silico solutions that take dose into consideration. A proposed framework for the integration of in silico and experimental information is provided along with a case study describing how computational methods have been used to successfully respond to a regulatory question concerning non-genotoxic impurities in chemically synthesized pharmaceuticals.

Trade-off Predictivity and Explainability for Machine-Learning Powered Predictive Toxicology: An in-Depth Investigation with Tox21 Data Sets

Selecting a model in predictive toxicology often involves a trade-off between prediction performance and explainability: should we sacrifice the model performance to gain explainability or vice versa. Here we present a comprehensive study to assess algorithm and feature influences on model performance in chemical toxicity research. We conducted over 5000 models for a Tox21 bioassay data set of 65 assays and ∼7600 compounds. Seven molecular representations as features and 12 modeling approaches varying in complexity and explainability were employed to systematically investigate the impact of various factors on model performance and explainability. We demonstrated that end points dictated a model's performance, regardless of the chosen modeling approach including deep learning and chemical features. Overall, more complex models such as (LS-)SVM and Random Forest performed marginally better than simpler models such as linear regression and KNN in the presented Tox21 data analysis. Since a simpler model with acceptable performance often also is easy to interpret for the Tox21 data set, it clearly was the preferred choice due to its better explainability. Given that each data set had its own error structure both for dependent and independent variables, we strongly recommend that it is important to conduct a systematic study with a broad range of model complexity and feature explainability to identify model balancing its predictivity and explainability.

DeepDILI: Deep Learning-Powered Drug-Induced Liver Injury Prediction Using Model-Level Representation

Drug-induced liver injury (DILI) is the most frequently reported single cause of safety-related withdrawal of marketed drugs. It is essential to identify drugs with DILI potential at the early stages of drug development. In this study, we describe a deep learning-powered DILI (DeepDILI) prediction model created by combining model-level representation generated by conventional machine learning (ML) algorithms with a deep learning framework based on Mold2 descriptors. We conducted a comprehensive evaluation of the proposed DeepDILI model performance by posing several critical questions: (1) Could the DILI potential of newly approved drugs be predicted by accumulated knowledge of early approved ones? (2) is model-level representation more informative than molecule-based representation for DILI prediction? and (3) could improved model explainability be established? For question 1, we developed the DeepDILI model using drugs approved before 1997 to predict the DILI potential of those approved thereafter. As a result, the DeepDILI model outperformed the five conventional ML algorithms and two state-of-the-art ensemble methods with a Matthews correlation coefficient (MCC) value of 0.331. For question 2, we demonstrated that the DeepDILI model's performance was significantly improved (i.e., a MCC improvement of 25.86% in test set) compared with deep neural networks based on molecule-based representation. For question 3, we found 21 chemical descriptors that were enriched, suggesting a strong association with DILI outcome. Furthermore, we found that the DeepDILI model has more discrimination power to identify the DILI potential of drugs belonging to the World Health Organization therapeutic category of 'alimentary tract and metabolism'. Moreover, the DeepDILI model based on Mold2 descriptors outperformed the ones with Mol2vec and MACCS descriptors. Finally, the DeepDILI model was applied to the recent real-world problem of predicting any DILI concern for potential COVID-19 treatments from repositioning drug candidates. Altogether, this developed DeepDILI model could serve as a promising tool for screening for DILI risk of compounds in the preclinical setting, and the DeepDILI model is publicly available through https://github.com/TingLi2016/DeepDILI.

[The trends in predicting drug-induced liver injury]

Drug-induced liver injury (DILI) is the major reason for the discontinuation of new drug development and the withdrawal of drugs from the market. Hence, the evaluation systems which predict the onset of DILI in the pre-clinical stage are needed. To date, many researchers have conducted the mechanism of DILI, but the DILI prediction is poor because of the complexity of DILI. In this regard, based on the information obtained from basic research and clinical case, several pharmaceutical companies have been developed DILI prediction methods with high sensitivity and specificity by combining multiple targets. Another reason for low predictability is derived from the conventional culture method which causes a rapid decrease in hepatocyte function. To overcome these problems, the construction of a high-level in vitro evaluation system has been developed and applied to DILI evaluation. On the other hand, these in vitro evaluation methods require a lot of labor and cost so, in silico prediction methods have also been constructed in recent years. Based on this point, this article reviews the trends in DILI prediction systems in the non-clinical stage.

Mechanism-based integrated assay systems for the prediction of drug-induced liver injury

Drug-induced liver injury (DILI) can cause hepatic failure and result in drug withdrawal from the market. It has host-related and compound-dependent mechanisms. Preclinical prediction of DILI risk is very challenging and safety assessments based on animals inadequately forecast human DILI risk. In contrast, human-derived in vitro cell culture-based models could improve DILI risk prediction accuracy. Here, we developed and validated an innovative method to assess DILI risk associated with various compounds. Fifty-four marketed and withdrawn drugs classified as DILI risks of "most concern", "less concern", and "no concern" were tested using a combination of four assays addressing mitochondrial injury, intrahepatic lipid accumulation, inhibition of bile canalicular network formation, and bile acid accumulation. Using the inhibitory potencies of the drugs evaluated in these in vitro tests, an algorithm with the highest available DILI risk prediction power was built by artificial neural network (ANN) analysis. It had an overall forecasting accuracy of 73%. We excluded the intrahepatic lipid accumulation assay to avoid overfitting. The accuracy of the algorithm in terms of predicting DILI risks was 62% when it was constructed by ANN but only 49% when it was built by the point-added scoring method. The final algorithm based on three assays made no DILI risk prediction errors such as "most concern " instead of "no concern" and vice-versa. Our mechanistic approach may accurately predict DILI risks associated with numerous candidate drugs.

Gene Expression Data Based Deep Learning Model for Accurate Prediction of Drug-Induced Liver Injury in Advance

Drug-induced liver injury (DILI), one of the most common adverse effects, leads to drug development failure or withdrawal from the market in most cases, showing an emerging challenge that is to accurately predict DILI in the early stage. Recently, the vast amount of gene expression data provides us valuable information for distinguishing DILI on a genomic scale. Moreover, the deep learning algorithm is a powerful strategy to automatically learn important features from raw and noisy data and shows great success in the field of medical diagnosis. In this study, a gene expression data based deep learning model was developed to predict DILI in advance by using gene expression data associated with DILI collected from ArrayExpress and then optimized by feature gene selection and parameters optimization. In addition, the previous machine learning algorithm support vector machine (SVM) was also used to construct another prediction model based on the same data sets, comparing the model performance with the optimal DL model. Finally, the evaluation test using 198 randomly selected samples showed that the optimal DL model achieved 97.1% accuracy, 97.4% sensitivity, 96.8% specificity, 0.942 matthews correlation coefficient, and 0.989 area under the ROC curve, while the performance of SVM model only reached 88.9% accuracy, 78.8% sensitivity, 99.0% specificity, 0.794 matthews correlation coefficient, and 0.901 area under the ROC curve. Furthermore, external data sets verification and animal experiments were conducted to assess the optimal DL model performance. Finally, the predicted results of the optimal DL model were almost consistent with experiment results. These results indicated that our gene expression data based deep learning model could systematically and accurately predict DILI in advance. It could be a useful tool to provide safety information for drug discovery and clinical rational drug use in early stage and become an important part of drug safety assessment.

Development of quantitative structure-activity relationship models to predict potential nephrotoxic ingredients in traditional Chinese medicines

The broad use of traditional Chinese medicines (TCMs) and the accompanied incidences of kidney injury have attracted considerable interest in investigating the responsible toxic ingredients. It is challenging to evaluate toxicity of TCMs since they contain complex mixtures of phytochemicals. Quantitative structure-activity relationship (QSAR) is an efficient tool to predict toxicity and QSAR study on TCMs-induced nephrotoxicity remains lacked. We developed QSAR models using three datasets of 609 compounds: natural products, drugs, and mixed (contained both kinds of data) datasets. Each dataset was used for modelling by utilizing artificial neural networks (ANN) and support vector machines (SVM) algorithms separately. Both internal and external validations were performed on each model. Six QSAR models were developed and yielded reliable performance in the internal validation. For external validation, 30 ingredients in the TCMs were predicted well by the natural product models (accuracy: ANN 96.7%, SVM 93.3%). The mixed models (accuracy: ANN 76.7%, SVM 66.7%) showed a better performance than the drug models (accuracy: ANN 50%, SVM 53.3%). Particularly, natural product models produced the most reliable results. It has the application not only on screening the nephrotoxic ingredients in TCMs, but it is also helpful at prioritizing the subsequent toxicity testing of natural products.

Advancing Predictive Hepatotoxicity at the Intersection of Experimental, in Silico, and Artificial Intelligence Technologies

Adverse drug reactions, particularly those that result in drug-induced liver injury (DILI), are a major cause of drug failure in clinical trials and drug withdrawals. Hepatotoxicity-mediated drug attrition occurs despite substantial investments of time and money in developing cellular assays, animal models, and computational models to predict its occurrence in humans. Underperformance in predicting hepatotoxicity associated with drugs and drug candidates has been attributed to existing gaps in our understanding of the mechanisms involved in driving hepatic injury after these compounds perfuse and are metabolized by the liver. Herein we assess in vitro, in vivo (animal), and in silico strategies used to develop predictive DILI models. We address the effectiveness of several two- and three-dimensional in vitro cellular methods that are frequently employed in hepatotoxicity screens and how they can be used to predict DILI in humans. We also explore how humanized animal models can recapitulate human drug metabolic profiles and associated liver injury. Finally, we highlight the maturation of computational methods for predicting hepatotoxicity, the untapped potential of artificial intelligence for improving in silico DILI screens, and how knowledge acquired from these predictions can shape the refinement of experimental methods.