A computational toxicogenomics approach identifies a list of highly hepatotoxic compounds from a large microarray database

Artificial Intelligence in Drug Toxicity Prediction: Recent Advances, Challenges, and Future Perspectives

Toxicity prediction is a critical step in the drug discovery process that helps identify and prioritize compounds with the greatest potential for safe and effective use in humans, while also reducing the risk of costly late-stage failures. It is estimated that over 30% of drug candidates are discarded owing to toxicity. Recently, artificial intelligence (AI) has been used to improve drug toxicity prediction as it provides more accurate and efficient methods for identifying the potentially toxic effects of new compounds before they are tested in human clinical trials, thus saving time and money. In this review, we present an overview of recent advances in AI-based drug toxicity prediction, including the use of various machine learning algorithms and deep learning architectures, of six major toxicity properties and Tox21 assay end points. Additionally, we provide a list of public data sources and useful toxicity prediction tools for the research community and highlight the challenges that must be addressed to enhance model performance. Finally, we discuss future perspectives for AI-based drug toxicity prediction. This review can aid researchers in understanding toxicity prediction and pave the way for new methods of drug discovery.

Salecan confers anti-inflammatory effects in liver injury via regulating gut microbiota and its metabolites

Salecan, a natural β-glucan and one of the novel food ingredients approved in China, has been shown a variety of positive health effects, yet the mechanism of liver injury remains poorly understood. In addition, β-glucan could induce the shifts in gut microbiota, however, whether modulation of gut microbiota by β-glucan is associated with their positive health effects remain elusive. Here, the anti-inflammatory effects and the underlying mechanism of Salecan supplementation in CCl4-induced liver injury were investigated. After 8 weeks of treatment, we observed that Salecan alleviated liver injury by regulating inflammatory response and M2 macrophage polarization. In addition, Salecan treatment modulated the composition of gut microbiota and antibiotic cocktail treatment indicated that the hepatoprotective effect of Salecan was dependent on the gut microbiota. Fecal microbiota transplantation was used to further verify the mechanism, and we confirmed that microbial colonization partially alleviated liver injury. Besides, microbiota-derived metabolites of Salecan also contributed to the hepatoprotective and anti-inflammatory effect of Salecan against liver injury. These findings supported that Salecan intervention attenuated liver injury by regulating gut microbiota and its metabolites.

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.

Navigating Transcriptomic Connectivity Mapping Workflows to Link Chemicals with Bioactivities

Screening new compounds for potential bioactivities against cellular targets is vital for drug discovery and chemical safety. Transcriptomics offers an efficient approach for assessing global gene expression changes, but interpreting chemical mechanisms from these data is often challenging. Connectivity mapping is a potential data-driven avenue for linking chemicals to mechanisms based on the observation that many biological processes are associated with unique gene expression signatures (gene signatures). However, mining the effects of a chemical on gene signatures for biological mechanisms is challenging because transcriptomic data contain thousands of noisy genes. New connectivity mapping approaches seeking to distinguish signal from noise continue to be developed, spurred by the promise of discovering chemical mechanisms, new drugs, and disease targets from burgeoning transcriptomic data. Here, we analyze these approaches in terms of diverse transcriptomic technologies, public databases, gene signatures, pattern-matching algorithms, and statistical evaluation criteria. To navigate the complexity of connectivity mapping, we propose a harmonized scheme to coherently organize and compare published workflows. We first standardize concepts underlying transcriptomic profiles and gene signatures based on various transcriptomic technologies such as microarrays, RNA-Seq, and L1000 and discuss the widely used data sources such as Gene Expression Omnibus, ArrayExpress, and MSigDB. Next, we generalize connectivity mapping as a pattern-matching task for finding similarity between a query (e.g., transcriptomic profile for new chemical) and a reference (e.g., gene signature of known target). Published pattern-matching approaches fall into two main categories: vector-based use metrics like correlation, Jaccard index, etc., and aggregation-based use parametric and nonparametric statistics (e.g., gene set enrichment analysis). The statistical methods for evaluating the performance of different approaches are described, along with comparisons reported in the literature on benchmark transcriptomic data sets. Lastly, we review connectivity mapping applications in toxicology and offer guidance on evaluating chemical-induced toxicity with concentration-response transcriptomic data. In addition to serving as a high-level guide and tutorial for understanding and implementing connectivity mapping workflows, we hope this review will stimulate new algorithms for evaluating chemical safety and drug discovery using transcriptomic data.

Artificial Intelligence-Based Toxicity Prediction of Environmental Chemicals: Future Directions for Chemical Management Applications

Recently, research on the development of artificial intelligence (AI)-based computational toxicology models that predict toxicity without the use of animal testing has emerged because of the rapid development of computer technology. Various computational toxicology techniques that predict toxicity based on the structure of chemical substances are gaining attention, including the quantitative structure-activity relationship. To understand the recent development of these models, we analyzed the databases, molecular descriptors, fingerprints, and algorithms considered in recent studies. Based on a selection of 96 papers published since 2014, we found that AI models have been developed to predict approximately 30 different toxicity end points using more than 20 toxicity databases. For model development, molecular access system and extended-connectivity fingerprints are the most commonly used molecular descriptors. The most used algorithm among the machine learning techniques is the random forest, while the most used algorithm among the deep learning techniques is a deep neural network. The use of AI technology in the development of toxicity prediction models is a new concept that will aid in achieving a scientific accord and meet regulatory applications. The comprehensive overview provided in this study will provide a useful guide for the further development and application of toxicity prediction models.

Chemical toxicity prediction based on semi-supervised learning and graph convolutional neural network

As safety is one of the most important properties of drugs, chemical toxicology prediction has received increasing attentions in the drug discovery research. Traditionally, researchers rely on in vitro and in vivo experiments to test the toxicity of chemical compounds. However, not only are these experiments time consuming and costly, but experiments that involve animal testing are increasingly subject to ethical concerns. While traditional machine learning (ML) methods have been used in the field with some success, the limited availability of annotated toxicity data is the major hurdle for further improving model performance. Inspired by the success of semi-supervised learning (SSL) algorithms, we propose a Graph Convolution Neural Network (GCN) to predict chemical toxicity and trained the network by the Mean Teacher (MT) SSL algorithm. Using the Tox21 data, our optimal SSL-GCN models for predicting the twelve toxicological endpoints achieve an average ROC-AUC score of 0.757 in the test set, which is a 6% improvement over GCN models trained by supervised learning and conventional ML methods. Our SSL-GCN models also exhibit superior performance when compared to models constructed using the built-in DeepChem ML methods. This study demonstrates that SSL can increase the prediction power of models by learning from unannotated data. The optimal unannotated to annotated data ratio ranges between 1:1 and 4:1. This study demonstrates the success of SSL in chemical toxicity prediction; the same technique is expected to be beneficial to other chemical property prediction tasks by utilizing existing large chemical databases. Our optimal model SSL-GCN is hosted on an online server accessible through: https://app.cbbio.online/ssl-gcn/home .

Artificial Intelligence in Drug Discovery: A Comprehensive Review of Data-driven and Machine Learning Approaches

As expenditure on drug development increases exponentially, the overall drug discovery process requires a sustainable revolution. Since artificial intelligence (AI) is leading the fourth industrial revolution, AI can be considered as a viable solution for unstable drug research and development. Generally, AI is applied to fields with sufficient data such as computer vision and natural language processing, but there are many efforts to revolutionize the existing drug discovery process by applying AI. This review provides a comprehensive, organized summary of the recent research trends in AI-guided drug discovery process including target identification, hit identification, ADMET prediction, lead optimization, and drug repositioning. The main data sources in each field are also summarized in this review. In addition, an in-depth analysis of the remaining challenges and limitations will be provided, and proposals for promising future directions in each of the aforementioned areas.

Transcriptomics in Toxicogenomics, Part III: Data Modelling for Risk Assessment

Transcriptomics data are relevant to address a number of challenges in Toxicogenomics (TGx). After careful planning of exposure conditions and data preprocessing, the TGx data can be used in predictive toxicology, where more advanced modelling techniques are applied. The large volume of molecular profiles produced by omics-based technologies allows the development and application of artificial intelligence (AI) methods in TGx. Indeed, the publicly available omics datasets are constantly increasing together with a plethora of different methods that are made available to facilitate their analysis, interpretation and the generation of accurate and stable predictive models. In this review, we present the state-of-the-art of data modelling applied to transcriptomics data in TGx. We show how the benchmark dose (BMD) analysis can be applied to TGx data. We review read across and adverse outcome pathways (AOP) modelling methodologies. We discuss how network-based approaches can be successfully employed to clarify the mechanism of action (MOA) or specific biomarkers of exposure. We also describe the main AI methodologies applied to TGx data to create predictive classification and regression models and we address current challenges. Finally, we present a short description of deep learning (DL) and data integration methodologies applied in these contexts. Modelling of TGx data represents a valuable tool for more accurate chemical safety assessment. This review is the third part of a three-article series on Transcriptomics in Toxicogenomics.

Predictability of drug-induced liver injury by machine learning

Background

Drug-induced liver injury (DILI) is a major concern in drug development, as hepatotoxicity may not be apparent at early stages but can lead to life threatening consequences. The ability to predict DILI from in vitro data would be a crucial advantage. In 2018, the Critical Assessment Massive Data Analysis group proposed the CMap Drug Safety challenge focusing on DILI prediction.

Methods and results

The challenge data included Affymetrix GeneChip expression profiles for the two cancer cell lines MCF7 and PC3 treated with 276 drug compounds and empty vehicles. Binary DILI labeling and a recommended train/test split for the development of predictive classification approaches were also provided. We devised three deep learning architectures for DILI prediction on the challenge data and compared them to random forest and multi-layer perceptron classifiers. On a subset of the data and for some of the models we additionally tested several strategies for balancing the two DILI classes and to identify alternative informative train/test splits. All the models were trained with the MAQC data analysis protocol (DAP), i.e., 10x5 cross-validation over the training set. In all the experiments, the classification performance in both cross-validation and external validation gave Matthews correlation coefficient (MCC) values below 0.2. We observed minimal differences between the two cell lines. Notably, deep learning approaches did not give an advantage on the classification performance.

Discussion

We extensively tested multiple machine learning approaches for the DILI classification task obtaining poor to mediocre performance. The results suggest that the CMap expression data on the two cell lines MCF7 and PC3 are not sufficient for accurate DILI label prediction.

Reviewers

This article was reviewed by Maciej Kandula and Paweł P. Labaj.

Genome-Scale Characterization of Toxicity-Induced Metabolic Alterations in Primary Hepatocytes

Context-specific GEnome-scale metabolic Network REconstructions (GENREs) provide a means to understand cellular metabolism at a deeper level of physiological detail. Here, we use transcriptomics data from chemically-exposed rat hepatocytes to constrain a GENRE of rat hepatocyte metabolism and predict biomarkers of liver toxicity using the Transcriptionally Inferred Metabolic Biomarker Response algorithm. We profiled alterations in cellular hepatocyte metabolism following in vitro exposure to four toxicants (acetaminophen, carbon tetrachloride, 2,3,7,8-tetrachlorodibenzodioxin, and trichloroethylene) for six hour. TIMBR predictions were compared with paired fresh and spent media metabolomics data from the same exposure conditions. Agreement between computational model predictions and experimental data led to the identification of specific metabolites and thus metabolic pathways associated with toxicant exposure. Here, we identified changes in the TCA metabolites citrate and alpha-ketoglutarate along with changes in carbohydrate metabolism and interruptions in ATP production and the TCA Cycle. Where predictions and experimental data disagreed, we identified testable hypotheses to reconcile differences between the model predictions and experimental data. The presented pipeline for using paired transcriptomics and metabolomics data provides a framework for interrogating multiple omics datasets to generate mechanistic insight of metabolic changes associated with toxicological responses.

Chinese Herbal Medicine Hepatotoxicity: The Evaluation and Recognization Based on Large-scale Evidence Database

Background:

Due to the special nature of Chinese Herbal medicine and the complexity of its clinical use, it is difficult to identify and evaluate its toxicity and resulting herb induced liver injury (HILI).

Methods:

First, the database would provide full profile of HILI from the basic ingredients to clinical out-comes by the most advanced algorithms of artificial intelligence, and it is also possible that we can predict possibilities of HILI after patients taking Chinese herbs by individual patient evaluation and prediction. Second, the database would solve the chaos and lack of the relevant data faced by the current basic re-search and clinical practice of Chinese Herbal Medicine. Third, we can also screen the susceptible patients from the database and thus prevent the accidents of HILI from the very beginning.

Results:

The Roussel Uclaf Causality Assessment Method (RUCAM) is the most accepted method to evalu-ate DILI, but at present before using the RUCAM evaluation method, data resource collection and analysis are yet to be perfected. Based on existing research on drug-metabolizing enzymes mediating reactive me-tabolites (RMs), the aim of this study is to explore the possibilities and methods of building multidimen-sional hierarchical database composing of RMs evidence library, Chinese herbal evidence library, and indi-vidualized reports evidence library of herb induced liver injury HILI.

Conclusion:

The potential benefits lie in its ability to organize, use vast amounts of evidence and use big data mining techniques at the center for Chinese herbal medicine liver toxicity research, which is the most difficult key point of scientific research to be investigated in the next few years.

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.

Literature optimized integration of gene expression for organ-specific evaluation of toxicogenomics datasets

The study of drug toxicity in human organs is complicated by their complex inter-relations and by the obvious difficulty to testing drug effects on biologically relevant material. Animal models and human cell cultures offer alternatives for systematic and large-scale profiling of drug effects on gene expression level, as typically found in the so-called toxicogenomics datasets. However, the complexity of these data, which includes variable drug doses, time points, and experimental setups, makes it difficult to choose and integrate the data, and to evaluate the appropriateness of one or another model system to study drug toxicity (of particular drugs) of particular human organs. Here, we define a protocol to integrate drug-wise rankings of gene expression changes in toxicogenomics data, which we apply to the TG-GATEs dataset, to prioritize genes for association to drug toxicity in liver or kidney. Contrast of the results with sets of known human genes associated to drug toxicity in the literature allows to compare different rank aggregation approaches for the task at hand. Collectively, ranks from multiple models point to genes not previously associated to toxicity, notably, the PCNA clamp associated factor (PCLAF), and genes regulated by the master regulator of the antioxidant response NFE2L2, such as NQO1 and SRXN1. In addition, comparing gene ranks from different models allowed us to evaluate striking differences in terms of toxicity-associated genes between human and rat hepatocytes or between rat liver and rat hepatocytes. We interpret these results to point to the different molecular functions associated to organ toxicity that are best described by each model. We conclude that the expected production of toxicogenomics panels with larger numbers of drugs and models, in combination with the ongoing increase of the experimental literature in organ toxicity, will lead to increasingly better associations of genes for organism toxicity.

Evaluation of 12 mouse marker genes in rat toxicogenomics public data, Open TG-GATEs: Discrimination of genotoxic from non-genotoxic hepatocarcinogens

Previously, we proposed 12 marker genes (Aen, Bax, Btg2, Ccnf, Ccng1, Cdkn1a, Gdf15, Lrp1, Mbd1, Phlda3, Plk2 and Tubb4b) to discriminate mouse genotoxic hepatocarcinogens (GTHC) from non-genotoxic hepatocarcinogens (NGTHC). This was determined by qPCR and principal component analysis (PCA), as the aim of an in vivo short-term screening for genotoxic hepatocarcinogens. For this paper, we conducted an application study of the 12 mouse marker genes to rat data, Open TG-GATEs (public data). We analyzed five typical rat GTHC (2-acetamodofluorene, aflatoxin B1, 2-nitrofluorene, N-nitrosodiethylamine and N-nitrosomorpholine), and not only seven typical rat NGTHC (clofibrate, ethanol, fenofibrate, gemfibrozil, hexachlorobenzene, phenobarbital and WY-14643) but also 11 non-genotoxic non-hepatocarcinogens (NGTNHC; allyl alcohol, aspirin, caffeine, chlorpheniramine, chlorpropamide, dexamethasone, diazepam, indomethacin, phenylbutazone, theophylline and tolbutamide) from Open TG-GATEs. The analysis was performed at 3, 6, 9 and 24 h after a single administration and 4, 8, 15 and 29 days after repeated administrations. We transferred Open TG-GATEs DNA microarray data into log₂ data using the "R Project for Statistical Computing". GTHC-specific dose-dependent gene expression changes were observed and significance assessed with the Williams test. Similar significant changes were observed during 3-24 h and 4-29 days, assessed with Welch's t-test, except not for NGTHC or NGTNHC. Significant differential changes in gene expression were observed between GTHC and NGTHC in 11 genes (except not Tubb4b) and between GTHC and NGTNHC in all 12 genes at 24 h and 10 genes (except Ccnf and Mbd1) at 29 days, per Tukey's test. PCA successfully discriminated GTHC from NGTHC and NGTNHC at 24 h and 29 days. The results demonstrate that 12 previously proposed mouse marker genes are useful for discriminating rat GTHC from NGTHC and NGTNHC from Open TG-GATEs.

Network and Pathway Analysis of Toxicogenomics Data

Toxicogenomics is the study of the molecular effects of chemical, biological and physical agents in biological systems, with the aim of elucidating toxicological mechanisms, building predictive models and improving diagnostics. The vast majority of toxicogenomics data has been generated at the transcriptome level, including RNA-seq and microarrays, and large quantities of drug-treatment data have been made publicly available through databases and repositories. Besides the identification of differentially expressed genes (DEGs) from case-control studies or drug treatment time series studies, bioinformatics methods have emerged that infer gene expression data at the molecular network and pathway level in order to reveal mechanistic information. In this work we describe different resources and tools that have been developed by us and others that relate gene expression measurements with known pathway information such as over-representation and gene set enrichment analyses. Furthermore, we highlight approaches that integrate gene expression data with molecular interaction networks in order to derive network modules related to drug toxicity. We describe the two main parts of the approach, i.e., the construction of a suitable molecular interaction network as well as the conduction of network propagation of the experimental data through the interaction network. In all cases we apply methods and tools to publicly available rat in vivo data on anthracyclines, an important class of anti-cancer drugs that are known to induce severe cardiotoxicity in patients. We report the results and functional implications achieved for four anthracyclines (doxorubicin, epirubicin, idarubicin, and daunorubicin) and compare the information content inherent in the different computational approaches.

Latest advances in diagnosing and predicting DILI: what was new in 2017?

Drug-induced liver injury (DILI) remains an increasingly recognized cause of hepatotoxicity and liver failure worldwide. In 2017, we continued to learn about predicting, diagnosing, and prognosticating drug hepatotoxicity. Areas covered: In this review, we selected from over 1200 articles from 2017 to synopsize updates in DILI. There were new HLA haplotypes associated with medications including HLA-C0401 and HLA-B*14. There has been continued work with quantitative systems pharmacology, particularly with the DILIsym® initiative, which employs mathematical representations of DILI mechanisms to predict hepatotoxicity in simulated populations. Additionally, knowledge regarding microRNAs (miRNAs) continues to expand. Some new miRNAs this past year include miRNA-223 and miRNA-605. Aside from miRNAs, other biomarkers for diagnosis, prognosis, and even prediction of DILI were explored. Studies on K18, OPN, and MCSFR have correlated DILI and liver-associated death within 6 months. Conversely, a new prognostic panel using apolipoportein-A1 and haptoglobin has been proposed to predict recovery. Further study of CDH5 has also provided researchers a possible new biomarker for prediction and susceptibility to DILI. Expert commentary: Although research on DILI remains quite promising, there is yet to be a reliable, simple method to predict, diagnose, and risk assess this form of hepatotoxicity.

Transcriptional Responses Reveal Similarities Between Preclinical Rat Liver Testing Systems

Toxicogenomics (TGx) is an important tool to gain an enhanced understanding of toxicity at the molecular level. Previously, we developed a pair ranking (PRank) method to assess in vitro to in vivo extrapolation (IVIVE) using toxicogenomic datasets from the Open Toxicogenomics Project-Genomics Assisted Toxicity Evaluation System (TG-GATEs) database. With this method, we investiagted three important questions that were not addressed in our previous study: (1) is a 1-day in vivo short-term assay able to replace the 28-day standard and expensive toxicological assay? (2) are some biological processes more conservative across different preclinical testing systems than others? and (3) do these preclinical testing systems have the similar resolution in differentiating drugs by their therapeutic uses? For question 1, a high similarity was noted (PRank score = 0.90), indicating the potential utility of shorter term in vivo studies to predict outcome in longer term and more expensive in vivo model systems. There was a moderate similarity between rat primary hepatocytes and in vivo repeat-dose studies (PRank score = 0.71) but a low similarity (PRank score = 0.56) between rat primary hepatocytes and in vivo single dose studies. To address question 2, we limited the analysis to gene sets relevant to specific toxicogenomic pathways and we found that pathways such as lipid metabolism were consistently over-represented in all three assay systems. For question 3, all three preclinical assay systems could distinguish compounds from different therapeutic categories. This suggests that any noted differences in assay systems was biological process-dependent and furthermore that all three systems have utility in assessing drug responses within a certain drug class. In conclusion, this comparison of three commonly used rat TGx systems provides useful information in utility and application of TGx assays.