CoChat
Sign Up
Cited by 193Computational Drug Discovery MethodsMachine Learning in Materials ScienceAnimal testing and alternatives

Machine Learning Based Toxicity Prediction: From Chemical Structural Description to Transcriptome Analysis

Open access
Wu, Yunyi, Wang, Guanyu·2018·International Journal of Molecular Sciences
DOI 10.3390/ijms19082358PMC PMC6121588
Open this paper in CoChat — free accountOpen PDF

Abstract

Toxicity prediction is very important to public health. Among its many applications, toxicity prediction is essential to reduce the cost and labor of a drug’s preclinical and clinical trials, because a lot of drug evaluations (cellular, animal, and clinical) can be spared due to the predicted toxicity. In the era of Big Data and artificial intelligence, toxicity prediction can benefit from machine learning, which has been widely used in many fields such as natural language processing, speech recognition, image recognition, computational chemistry, and bioinformatics, with excellent performance. In this article, we review machine learning methods that have been applied to toxicity prediction, including deep learning, random forests, k-nearest neighbors, and support vector machines. We also discuss the input parameter to the machine learning algorithm, especially its shift from chemical structural description only to that combined with human transcriptome data analysis, which can greatly enhance prediction accuracy.

AI snapshot

Written from the paper’s full text.

TL;DR

This review surveys machine-learning approaches for toxicity prediction, covering algorithms, chemical descriptors, and the shift toward combining chemical structure with human transcriptome data. It argues that transcriptome-informed models can improve accuracy, while current toxicity datasets may still limit deep learning’s advantage over traditional methods.

Key findings

  • Machine learning methods applied to toxicity prediction include deep learning, random forests, k-nearest neighbors, and support vector machines; performance depends on datasets and computational representations.
  • Chemical-structure-only prediction has produced varied results: CNN-based toxicity prediction reached an average AUC of 0.846; DNN multi-task neural networks reached AUC 0.9 for NIH/3T3 toxicity QSAR versus 0.87 for random forests; MGE-CNN yielded classification accuracy up to 95.0% and regression R² 0.861 for acute oral toxicity.
  • Chronic toxicity prediction from chemical structure alone has been weaker in some cases: k-NN-based LOEL prediction had test-set R² only 0.54, while k-NN oral LOAEL prediction reached AUC up to 0.814.
  • Adding transcriptome data improved prediction in cited studies: SVM classification of chemicals using RNA-Seq from 3D-cultured human pluripotent stem cell-derived neural cells achieved average AUC 0.91; a variable-nearest neighbor model predicting chemical-induced gene expression changes achieved AUC > 0.7; RF using chemical structure, target-gene expression, and mutation data achieved AUC about 0.8263.

Methods

  • Design: narrative review of machine-learning methods for toxicity prediction, focusing on the evolution from chemical structural descriptors alone to combined chemical and human transcriptome analysis.
  • Data/sample: no primary dataset; the review discusses toxicity databases such as TOXNET, HSDB, ToxCast/Tox21, and gene-expression resources including CMap and GEO.
  • Main analysis: synthesizes algorithms (ANN, deep learning, RF, k-NN, SVM), descriptor types (molecular fingerprints such as MACCS and ECFP, graph-based and CNN-learned fragments, chemical properties), and reported performance metrics such as AUC and R².

Limitations

  • The paper states that currently available toxicity datasets are not sufficiently large to support deep learning algorithms in further improving accuracy, and notes that traditional methods such as SVM and RF sometimes have higher AUC values than deep learning.
  • The paper states that chemical-structure-based toxicity prediction is only the first step of drug evaluation, and that using median expression of drug target genes may ignore tissue specificity and differential toxic reactions.
  • The text makes evident that this is a non-systematic review without primary validation or meta-analysis, and that performance comparisons are drawn from heterogeneous datasets, descriptors, and model settings, limiting direct comparability.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/ijms19082358

Cites

155 of 155 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 58 cited sources checked · 0 supported · 0 flagged

  • Findingreasoning

    Traditional machine learning methods such as SVM and RF have higher AUC values than deep learning algorithms.

    “One sees that traditional machine learning methods such as SVM and RF have higher AUC values than deep learning algorithms.”
    Quote in the paperSubstantial claim
  • Interpretationnone given

    Statistical approaches such as linear regression, multivariate analysis, and early shallow neural network models are difficult to extract more abstract features and thus difficult to predict with high accuracy.

    “statistical approaches, such as linear regression analysis, multivariate analysis, and early shallow neural network models are difficult to extract more abstract features, and are thus difficult to predict with high accuracy.”
    Quote in the paperNot a substantial claim
  • Interpretationreasoning

    Combining chemical structural description with human transcriptome data analysis can greatly enhance toxicity prediction accuracy.

    “We also discuss the input parameter to the machine learning algorithm, especially its shift from chemical structural description only to that combined with human transcriptome data analysis, which can greatly enhance prediction accuracy.”
    • Also stated in: § 7.3. Toxicity Prediction · hedged
      “Incorporating genetic information would render more accurate toxicity prediction and QSAR construction [133].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [133] Yamane, 2016, Prediction of developmental chemical toxicity based on gene networks of human embryonic stem cells, Nucleic Acids Res. source

      Not checked in the free run.

  • Premisecitation

    Animal experiments offer little guidance to human toxicity reactions due to inter-species differences and differential disease models.

    “the results offer little guidance to human toxicity reactions, due to inter-species differences and differential disease models [6,7].”
    • Also stated in: § 1. Introduction · plain
      “animal experiments cannot tell the human body’s response to new drugs and offer no risk exemption [6,9].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [6] Johnson, 2013, Fusion of nonclinical and clinical data to predict human drug safety, Expert Rev. Clin. Pharmacol. source

      Not checked in the free run.

    • cites [7] Akhtar, 2015, The Flaws and Human Harms of Animal Experimentation, Camb. Q. Healthc. Ethics source

      Not checked in the free run.

    • cites [9] Thomas, 2018, The US Federal Tox21 Program: A strategic and operational plan for continued leadership, Altex source

      Not checked in the free run.

  • Premisecitation

    Knowledge-based systems are difficult to complete highly automated work with high volume of data because maintaining completeness of knowledge bases becomes difficult as data increases.

    “with the amount of data increasing explosively, it becomes more and more difficult to maintain completeness of knowledge bases; thus, knowledge-based systems are difficult to complete highly automated work with high volume of data [18].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [18] Compton, 2000, Knowledge Based Systems That Have Some Idea of Their Limits, CIO

      Not checked in the free run.

  • May combine two claims. 30
  • Classified as probably not a substantial claim. 24
  • The quote only partly states this claim. 2

  • May combine two claims.

    Premisecitation

    Sitaxentan caused no explicit liver injury in animal experiments, whereas hepatotoxicity was prominent in humans.

    “Sitaxentan caused no explicit liver injury in animal experiments [8], whereas the hepatotoxicity was prominent in humans [4,5].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [8] Owen, 2012, An overview of the preclinical toxicity and potential carcinogenicity of sitaxentan (Thelin®), a potent endothelin receptor antagonist developed for pulmonary arterial hypertension, Regul. Toxicol. Pharmacol. source

      Not checked in the free run.

    • cites [4] Erve, 2013, Bioactivation of sitaxentan in liver microsomes, hepatocytes, and expressed human P450s with characterization of the glutathione conjugate by liquid chromatography tandem mass spectrometry, Chem. Res. Toxicol. source

      Not checked in the free run.

    • cites [5] Hoeper, 2011, Liver toxicity of sitaxentan in pulmonary arterial hypertension, Eur. Heart J.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    SVMs and RFs are both based on statistics; they thus perform well in structured and denser datasets.

    “SVMs and RFs are both based on statistics; they thus perform well in structured and denser datasets.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    RNN is more suitable for prediction or recognition of sequences; CNN is more suitable for recognition of spatial arrangement features.

    “The former is more suitable for prediction or recognition of sequences, such as natural language processing [59] and time series prediction [60,61]. The latter is more suitable for the recognition of spatial arrangement features, such as the shapes in graphics and images [62].”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [59] Cho, K., van Merrienboer, B., Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H., and Bengio, Y. (2014). Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation. arXiv. source

      Not checked in the free run.

    • cites [60] Che, 2018, Recurrent Neural Networks for Multivariate Time Series with Missing Values, Sci. Rep. source

      Not checked in the free run.

    • cites [61] Madhavan, P.G. (1993, January 31). Recurrent neural network for time series prediction. Proceedings of the 15th Annual International Conference of the IEEE Engineering in Medicine and Biology Societ, San Diego, CA, USA.

      Not checked in the free run.

    • cites [62] Jaderberg, M., Simonyan, K., Zisserman, A., and Kavukcuoglu, K. (2015). Spatial Transformer Networks. arXiv.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The ReLU activation function solved the vanishing gradient problem inherent of the sigmoid function.

    “the ReLU (Rectified Linear Unit) activation function was first proposed [43], which solved the vanishing gradient problem inherent of the sigmoid function.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [43] Geoffrey, 2011, Deep Sparse Rectifier Neural Networks, Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Deep learning excels when the working data are unstructured, sparse, and large.

    “Deep learning excels when the working data are unstructured, sparse, and large.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    Different machine learning methods perform differently; datasets and computational representations can significantly affect performance.

    “Different machine learning methods perform differently. Factors such as datasets and computational representations can significantly affect the performance.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    BP with sigmoid activation function has the vanishing gradient problem and is thus difficult to follow deeper and more abstract training.

    “it was pointed out that BP with sigmoid activation function has the vanishing gradient problem and is thus difficult to follow deeper and more abstract training [42].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [42] Hochreiter, 1998, The Vanishing Gradient Problem during Learning Recurrent Neural Nets and Problem Solutions, Int. J. Uncertain. Fuzz. Knowl.-Based Syst. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Whole genome expression, or transcriptome expression, reflects the state changes of a cell, either in vivo or in vitro.

    “Whole genome expression, or transcriptome expression, reflects the state changes of a cell, either in vivo or in vitro.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationreasoning

    Because drugs are designed for humans, toxicity testing on human cells is both clinically relevant and cost effective.

    “Because drugs are designed for humans, toxicity testing on human cells is both clinically relevant and cost effective.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Chavan et al. classified LD50 values using k-NN and predicted LOEL to measure chronic toxicity, with R square of test set only 0.54.

    “Chavan et al. classified the LD50 values of compounds using k-NN. Based on the classification, they predicted the LOEL (lowest observed effect level), which was then used to measure chronic toxicity. The R square value of the test set was only 0.54, however [93].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [93] Chavan, 2015, Acute Toxicity-Supported Chronic Toxicity Prediction: A k-Nearest Neighbor Coupled Read-Across Strategy, Int. J. Mol. Sci. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Chronic toxicity is more latent and hard to discover than acute toxicity.

    “When compared with acute toxicity, chronic toxicity is more latent and hard to discover.”
    Quote in the paperNot a substantial claim

  • May combine two claims. The quote only partly states this claim.

    Premisecitation

    Liu et al. found that RF and DNN performed better on the global dataset, while k-NN performed better on the unbalanced acute toxicity datasets, using ECFP uniformly.

    “It was found that RF and DNN performed better on the global dataset, while k-NN performed better on the unbalanced acute toxicity datasets. This result also highlights the importance of neighbor information in acute toxicity prediction [126].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [126] Liu, 2018, Assessing Deep and Shallow Learning Methods for Quantitative Prediction of Acute Chemical Toxicity, Toxicol. Sci. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    Molecular fingerprints encode chemical structures in great detail, which may sometimes be unnecessary or even disadvantageous.

    “Molecular fingerprints encode chemical structures in great detail (every atom or bond), which may sometimes be unnecessary or even disadvantageous (complicated and inefficient).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    If a cell has a high expression of a proto-oncogene, then the chance is high of the cell's carcinogenesis.

    “if a cell has a high expression of a proto-oncogene, then the chance is high of the cell’s carcinogenesis.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Schwartz et al. used toxic and non-toxic compounds to treat 3D-cultured human pluripotent stem cell-derived neural cells, RNA-Seq, and SVM to classify chemicals by toxicity, gaining average AUC of 0.91.

    “Schwartz et al. used both toxic and non-toxic compounds to treat 3D-cultured human pluripotent stem cell-derived neural cells, then used RNA-Seq to determine the whole genome expression profile, and then used SVM to classify the chemicals according to their toxicity. The scheme gained an average AUC value of 0.91 [132].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [132] Schwartz, 2015, Human pluripotent stem cell-derived neural constructs for predicting neural toxicity, Proc. Natl. Acad. Sci. USA source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Wu et al. improved traditional molecular descriptors using ESPH and auxiliary descriptors, and achieved the highest degree of fitness and accuracy with RF, Gradient Boosting Decision Tree, single-task deep learning, and multi-task deep learning.

    “Wu et al. recently improved traditional molecular descriptors using element specific persistent homology (ESPH) and auxiliary descriptors, where ESPH includes topological information from intermolecular interactions and homology analysis on each component of molecules. On this basis, they performed RF, Gradient Boosting Decision Tree, single-task deep learning, multi-task deep learning, multi-task”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [118] Wu, 2018, Quantitative toxicity prediction using topology based multi-task deep neural networks, J. Chem. Inform. Model. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Different combinations of chemical descriptors and machine learning models might perform differently.

    “Different combinations of chemical descriptors and machine learning models might perform differently.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Lu et al. used k-NN combined linear regression model to predict acute oral toxicity in rats and achieved an R square value of 0.712, using local chemical structure represented by molecular fingerprints.

    “in 2011, Lu et al. used k-NN combined linear regression model to predict acute oral toxicity in rats and achieved a R square value of 0.712, for which they utilized the local chemical structure that was represented by molecular fingerprints [31].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [31] Lu, 2014, Estimation of acute oral toxicity in rat using local lazy learning, J. Cheminform. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Martin et al. used global hierarchical clustering to predict acute toxicity of pesticides and obtained better results than linear regression.

    “Martin et al. used the global hierarchical clustering method to predict acute toxicity of pesticides and obtained better results than linear regression [125].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [125] Martin, 2017, Prediction of pesticide acute toxicity using two-dimensional chemical descriptors and target species classification, SAR QSAR Environ. Res. source

      Not checked in the free run.

  • May combine two claims. The quote only partly states this claim.

    Premisecitation

    Xu et al. used MGE-CNN to process standard molecular structure data and obtained a deep-minded fingerprint, yielding classification accuracy up to 95.0% and regression R square value of 0.861.

    “Xu et al. used an enhanced molecular graph encoding convolutional neural networks (MGE-CNN) (the gray box in Figure 3) to process the standard molecular structure data, and finally obtained the fingerprint. The fingerprint was further mined both forwardly and backwardly, which yielded the deep-minded fingerprint (the array of black dots in Figure 3). The deep-minded fingerprint was then tested by”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [127] Xu, 2017, Deep Learning Based Regression and Multiclass Models for Acute Oral Toxicity Prediction with Automatic Chemical Feature Extraction, J. Chem. Inf. Model. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    RNA-Seq technology has led to an explosion in data, which helps researchers gain a deeper understanding of biological mechanisms of toxic compounds, such as Benzo[a]pyrene, and better characterize harmful effects caused by chemicals.

    “Deep-sequencing RNA-Seq technology has led to an unrivaled explosion in the amount of data, which would help researchers to gain a deeper understanding of biological mechanisms (e.g., changes of cellular signaling pathways) of toxic compounds, such as Benzo[a]pyrene. This would, in turn, help researchers to better characterize harmful effects that are caused by chemicals [130].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [130] Van, 2012, RNA-Seq provides new insights in the transcriptome responses induced by the carcinogen benzo[a]pyrene, Br. J. Dermatol.

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Deep learning multi-task neural networks achieved an AUC of 0.9 for toxicity QSAR prediction of NIH/3T3 cells, slightly higher than the AUC of 0.87 in random forests, with molecular fingerprints as input.

    “Deep learning multi-task neural networks worked so well that the AUC value for toxicity QSAR prediction of NIH/3T3 cells (mouse embryonic fibroblast) can reach 0.9, which is slightly higher than the AUC of 0.87 in random forests, in which molecular fingerprints as input of the model [116].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [116] Dahl, G.E., Jaitly, N., and Salakhutdinov, R. (2014). Multi-task Neural Networks for QSAR Predictions. arXiv.

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The reason traditional ML methods have higher AUC than deep learning might be that currently available toxicity datasets are not sufficiently large to support deep learning algorithms to further improve their accuracy.

    “The reason might be that currently available toxicity datasets are not sufficiently large to support deep learning algorithms to further improve their accuracy.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Accuracy of toxicity prediction depends on the amount of data being collected.

    “Accuracy of toxicity prediction depends on the amount of data being collected.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    Chemical structure based toxicity prediction is only the first step of drug evaluation; subsequent steps include cell, animal, and clinical toxicity tests.

    “Chemical structure based toxicity prediction is only the first step of drug evaluation. The subsequent steps include cell, animal, and clinical toxicity tests.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Li et al. used RF, SVM, and k-NN to predict oral LOAEL of rats; k-NN obtained best performance with AUC up to 0.814.

    “In 2017, Li et al. used machine learning models, such as RF, SVM, and k-NN to predict the oral LOAEL (lowest observed adverse effect level) of rats. The method k-NN obtained the best performance, yielding AUC values up to 0.814 [128].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [128] Li, 2017, In silico prediction of chronic toxicity with chemical category approaches, RSC Adv. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationcitation

    Machine learning techniques may provide much more information about toxicity than ever before.

    “Together with “Big Data” science [159], machine learning techniques may provide much more information about toxicity than ever before.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [159] Zhu, 2014, Big data in chemical toxicity research: The use of high-throughput screening assays to identify potential toxicants, Chem. Res. Toxicol. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Because gene expression is not immediate, toxicity onset is often delayed and difficult to detect immediately after induction.

    “Because gene expression is not immediate, toxicity onset is often delayed and it is difficult to detect immediately after the induction.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Calculation of median expression of drug target genes may ignore tissue specificity and differential toxic reactions; tissue transcriptome data might be more specific and help extract more relevant features.

    “Calculation of median expression of drug target genes is useful, but they may ignore tissue specificity and differential toxic reactions. For example, a toxic drug may induce high expression of a particular gene only in the liver, but not in the other organs or tissues. The median value of the gene expression, being based on the whole body measurement, is thus very low and cannot reflect the drug’”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    In silico prediction is low-cost.

    “Secondly, in silico prediction is risk-free, low-costly, and of high throughput.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    One can not only distinguish between toxic and non-toxic, but also perform classified toxicity prediction (neurotoxins, carcinogens, etc.).

    “One can not only distinguish between toxic and non-toxic, but also perform classified toxicity prediction (neurotoxins, carcinogens, etc.).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Gayvert et al. performed classified toxicity prediction on FDA-approved drugs and failed drugs with RF, using chemical structure, median expression of drug targeted genes, and functional mutations, obtaining AUC about 0.8263.

    “Gayvert et al. performed classified toxicity prediction on FDA-approved drugs and drugs that had failed to pass toxicity-tests, with the RF supervised learning algorithm. The learning was from multiple sources: chemical structure characterizations, the median value of the expression of the drug targeted genes from the transcriptome of various tissues, the frequency or possibility of functional”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [157] Gayvert, 2016, A Data-Driven Approach to Predicting Successes and Failures of Clinical Trials, Cell Chem. Biol. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    Compared with random forest, deep learning can handle higher throughput and larger amounts of data, and deal with higher-level and more abstract features, resulting in better performance after subsequent data accumulation.

    “When compared with the random forest approach, the deep learning approach can handle higher throughput and larger amounts of data, and be capable to deal with higher-level and more abstract features, resulting in a better performance after subsequent data accumulation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Accurate prediction of late-onset chemical toxicity might be ascribed to the analysis of gene interaction networks: alterations caused by a compound propagate through gene-gene interactions, and chain reactions lead to genome instability and cytotoxicity.

    “Therefore, the accurate prediction of late-onset chemical toxicity might be ascribed to the analysis of gene interaction networks: alterations that are caused by a compound propagate through gene-gene interactions; and, the chain reactions finally lead to genome instability and cytotoxicity.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Stacked Denoising Autoencoder allows discovery of gene interaction patterns and specific gene expression patterns by extracting features from RNA-Seq data.

    “the dimension of RNA-Seq data can be reduced by techniques, such as Stacked Denoising Autoencoder (SDAE), which allows for the discovery of gene interaction patterns [154] and specific gene expression patterns [155] by extracting features from RNA-Seq data by a supervised learning classification model.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [154] Danaee, 2016, A Deep Learning Approach For Cancer Detection and Relevant Gene Identification, Pac. Symp. Biocomput.

      Not checked in the free run.

    • cites [155] Sharifi-Noghabi, H., Liu, Y., Erho, N., Shrestha, R., Alshalalfa, M., Davicioni, E., Collins, C.C., and Ester, M. (2018). Deep Genomic Signature for early metastasis prediction in prostate cancer. bioRxiv. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Yamane et al. used chemicals to treat human embryonic stem cells, analyzed transcriptomes, classified chemicals into categories, analyzed gene interaction networks, greatly improved accuracy of toxicity prediction and allowed prediction of delayed chemical toxicity with SVM.

    “Yamane et al. used chemicals to treat human embryonic stem cells and analyzed their transcriptomes. By classifying the chemicals into different categories, such as neurotoxins, genotoxic carcinogens, and non-genotoxic carcinogens, and by analyzing gene interaction networks, they gained much richer information, which greatly improved the accuracy of toxicity prediction and even allowed for them to”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [133] Yamane, 2016, Prediction of developmental chemical toxicity based on gene networks of human embryonic stem cells, Nucleic Acids Res. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    With gene expression data and digitalized chemical structures, one can use machine learning models to find deeper intrinsic links between the two, either from structure to effect or from effect to structure; the former helps QSAR prediction including toxicity, the latter helps design inducing drugs based on desired gene expression changes.

    “one can use the system to find deeper and intrinsic links between the two through machine learning models (Figure 5), by either establishing the association with chemical structures as input and gene expressions as output (from structure to effect), or vice versa (from effect to structure). The former can help with QSAR prediction, including toxicity, while the latter can help with the design of”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    In silico prediction is high-throughput.

    “Secondly, in silico prediction is risk-free, low-costly, and of high throughput.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationnone given

    Gene Ontology-based vector representation directly reflects gene functions and indirectly reflects gene interactions.

    “This representation directly reflects gene functions and indirectly reflects gene interactions.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Delayed toxicity is associated with changes in gene expression, which can affect the expression of downstream genes.

    “What underlay their success was the fact that delayed toxicity is associated with changes in gene expression, which can, in turn, affect the expression of downstream genes [134,135].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [134] Ippolito, 2016, Gene Expression Patterns Associated With Histopathology in Toxic Liver Fibrosis, Toxicol. Sci. source

      Not checked in the free run.

    • cites [135] Smith, 2016, Tumor Regression and Delayed Onset Toxicity Following B7-H4 CAR T Cell Therapy, Mol. Therapy J. Am. Soc. Gene Therapy source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    In silico prediction is risk-free.

    “Secondly, in silico prediction is risk-free, low-costly, and of high throughput.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Liu et al. established a variable-nearest neighbor model to predict QSAR between chemical structures and gene expression profiles, obtaining AUC >0.7.

    “Liu et al. established a variable-nearest neighbor model to predict the QSAR between chemical structures and gene expression profiles, and obtained an AUC value of more than 0.7 [139].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [139] Liu, 2017, Molecular Structure-Based Large-Scale Prediction of Chemical-Induced Gene Expression Changes, J. Chem. Inform. Model. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Many harmful and risky animal or clinical trials can be spared because toxicity is predicted by computers.

    “Firstly, many harmful and risky animal or clinical trials can be spared, due to toxicity predicted by computers.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    CMap is the most popular database to analyze the relationship between transcriptome data and drugs.

    “Among the databases, CMap is the most popular one to analyze the relationship between transcriptome data and drugs [140].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [140] Lamb, 2006, The Connectivity Map: Using gene-expression signatures to connect small molecules, genes, and disease, Science source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Vectors constructed based on similarity of different gene annotations in Gene Ontology allow quantification of similarities between genes.

    “In another method, vectors are constructed based on a similarity of different gene annotations in Gene Ontology, which allows for the quantification of similarities between genes [153].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [153] Duong, D., Eskin, E., and Li, J. (2017). A novel Word2vec based tool to estimate semantic similarity of genes by using Gene Ontology terms. bioRxiv.

      Not checked in the free run.

45 of 45

  • The paper relies on this without support.

    Information for toxicity prediction is primarily from the drug compound's chemical structure.

    “Information for toxicity prediction is primarily from the drug compound’s chemical structure.”

  • The paper relies on this without support.

    Asymmetric data are ubiquitous in the toxicity database because non-toxic compounds are not specifically labeled in the database.

    “Asymmetric data are ubiquitous in the toxicity database, because non-toxic compounds are not specifically labeled in the database.”

  • The paper relies on this without support.

    Toxicity prediction is essential to reduce the cost and labor of a drug’s preclinical and clinical trials because many drug evaluations can be spared due to predicted toxicity.

    “toxicity prediction is essential to reduce the cost and labor of a drug’s preclinical and clinical trials, because a lot of drug evaluations (cellular, animal, and clinical) can be spared due to the predicted toxicity.”

  • The paper relies on this without support.

    Human transcriptome data are often used in machine learning based toxicity prediction.

    “because human transcriptome data are often used”
4 of 4

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Acute-toxicity study dated 2011 in the text but 2014 in the reference

    Minor

    The text dates the cited Lu et al. study to 2011, while reference 31 lists the corresponding acute-toxicity study as published in 2014.

1 of 1

Cites

Flagged first, then by citations

1 of the 156 references couldn’t be matched to a known paper.

Research works printed but not linked: 11

  1. Janodia, 2007, Drug Development Process: A review, Pharm. Rev., 5, 2214
  2. Compton, 2000, Knowledge Based Systems That Have Some Idea of Their Limits, CIO, 15, 57
  3. Mitchell, T.M. (1997). Machine Learning, McGraw Hill.
  4. Bishop, C.M. (2006). Pattern Recognition and Machine Learning, Springer. [1st ed.].
  5. Srinivasan, A., and King, R.D. (1999). Using Inductive Logic Programming to construct Structure-Activity Relationshipsp, AAAI.
  6. Tin Kam, H. (1995, January 14–16). Random Decision Forests. Proceedings of the 3rd International Conference on Document Analysis and Recognition, Montreal, QC, Canada.
  7. Zahangir Alom, M., Taha, T.M., Yakopcic, C., Westberg, S., Hasan, M., Van Esesn, B.C., Awwal, A.A.S., and Asari, V.K. (2018). The History Began from AlexNet: A Comprehensive Survey on Deep Learning Approaches. arXiv.
  8. Jaderberg, M., Simonyan, K., Zisserman, A., and Kavukcuoglu, K. (2015). Spatial Transformer Networks. arXiv.
  9. Kumar, 2014, An Efficient Approach for Asymmetric Data Classification, Int. J. Innov. Res. Adv. Eng., 1, 157
  10. Dahl, G.E., Jaitly, N., and Salakhutdinov, R. (2014). Multi-task Neural Networks for QSAR Predictions. arXiv.
  11. Duong, D., Eskin, E., and Li, J. (2017). A novel Word2vec based tool to estimate semantic similarity of genes by using Gene Ontology terms. bioRxiv.

Cited by

Sorted by citations

Related

Authors
Wu, Y., & Wang, G.
Affiliations
Southern University of Science and Technology (CN)
Corresponding author
Wang, Guanyu
ORCID
Wang, Guanyu
Venue
International Journal of Molecular Sciences
Publisher
MDPI AG
Published
2018-08-10
Volume
19
Issue
8
Pages
2358
ISSN
1422-0067 (ISSN-L)
DOI
10.3390/ijms19082358
PMC
PMC6121588
OpenAlex
W2887896328
Semantic Scholar
60b5e9e616fde8a50bb016c8b7ecf95e0ab3a6f5
Cited by
193
References
156
Citations by year
2026: 16 · 2025: 25 · 2024: 27 · 2023: 27 · 2022: 33 · 2021: 32 · 2020: 19 · 2019: 13 · 2018: 1
FWCI
1.28
Type
journal article
Language
en
Open access
Licence
Locations
  • International Journal of Molecular Sciences (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Natural Science Foundation of China (61773196, 61471186)
  • Shenzhen Municipal Research Fund (JCYJ20170307104535585, JCYJ20170817104740861)
  • Shenzhen Peacock Plan (KQTD2016053117035204)
Topics
  • Physical Sciences › Computer Science › Computational Theory and Mathematics › Computational Drug Discovery Methods
  • Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
  • Health Sciences › Veterinary › Small Animals › Animal testing and alternatives
Keywords
toxicity prediction, machine learning, drug toxicity, transcriptomic data, deep learning, random forest, k-nearest neighbors, support vector machine, preclinical drug testing, computational toxicology
MeSH terms
Machine Learning, Animals, Humans, Natural Language Processing, Gene Expression Profiling, Transcriptome

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review