CoChat
Sign Up
Cited by 212Computational Drug Discovery MethodsMachine Learning in Materials ScienceAnimal testing and alternatives

The DOI matches its Crossref record; 90 of the 90 references it cites were found in a registry or an index.

eToxPred: a machine learning-based approach to estimate the toxicity of drug candidates

Open access
Pu, Limeng, Naderi, Misagh, Liu, Tairan, Wu, Hsiao-Chun, Mukhopadhyay, Supratik, Brylinski, Michal·2019·BMC Pharmacology and Toxicology
DOI 10.1186/s40360-018-0282-6PMC PMC6325674
Open this paper in CoChat — free accountOpen PDF

Abstract

BACKGROUND: The efficiency of drug development defined as a number of successfully launched new pharmaceuticals normalized by financial investments has significantly declined. Nonetheless, recent advances in high-throughput experimental techniques and computational modeling promise reductions in the costs and development times required to bring new drugs to market. The prediction of toxicity of drug candidates is one of the important components of modern drug discovery. RESULTS: In this work, we describe eToxPred, a new approach to reliably estimate the toxicity and synthetic accessibility of small organic compounds. eToxPred employs machine learning algorithms trained on molecular fingerprints to evaluate drug candidates. The performance is assessed against multiple datasets containing known drugs, potentially hazardous chemicals, natural products, and synthetic bioactive compounds. Encouragingly, eToxPred predicts the synthetic accessibility with the mean square error of only 4% and the toxicity with the accuracy of as high as 72%. CONCLUSIONS: eToxPred can be incorporated into protocols to construct custom libraries for virtual screening in order to filter out those drug candidates that are potentially toxic or would be difficult to synthesize. It is freely available as a stand-alone software at https://github.com/pulimeng/etoxpred .

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1186/s40360-018-0282-6

Cites

90 of 90 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Schematics of various machine learning classifiers. (a) A two-layered Boltzmann Machine with 3 hidden nodes h and 2 visible nodes v. Nodes are fully connected. (b) A Restricted Boltzmann Machine (RBM) with the same nodes as in A. Nodes belonging to the same layer are not connected. (c) A Deep Belief Network with a visible layer V and 3 hidden layers H. Individual layers correspond to RBMs that are stacked against one another. (d) A Random Forest with 3 trees T. For a given instance, each tree predicts a class based on a subset of the input set. The final class assignment is obtained by the majority voting of individual trees
Fig. 1 Schematics of various machine learning classifiers. (a) A two-layered Boltzmann Machine with 3 hidden nodes h and 2 visible nodes v. Nodes are fully connected. (b) A Restricted Boltzmann Machine (RBM) with the same nodes as in A. Nodes belonging to the same layer are not connected. (c) A Deep Belief Network with a visible layer V and 3 hidden layers H. Individual layers correspond to RBMs that are stacked against one another. (d) A Random Forest with 3 trees T. For a given instance, each tree predicts a class based on a subset of the input set. The final class assignment is obtained by the majority voting of individual treesCC BY · PubMed Central
Fig. 2 Resampling strategy to balance the dataset. The histogram shows the distribution of SAscore values across the training set before (solid gray line) and after (dashed black line) the over/under-sampling
Fig. 2 Resampling strategy to balance the dataset. The histogram shows the distribution of SAscore values across the training set before (solid gray line) and after (dashed black line) the over/under-samplingCC BY · PubMed Central
Fig. 3 SAscore prediction for several datasets. The scatter plot shows the correlation between the predicted and true SAscore values for active compounds from the Directory of Useful Decoys, Extended (DUD-E), FDA-approved drugs, and natural products from the NuBBE and UNPD databases. The regression line is dashed black
Fig. 3 SAscore prediction for several datasets. The scatter plot shows the correlation between the predicted and true SAscore values for active compounds from the Directory of Useful Decoys, Extended (DUD-E), FDA-approved drugs, and natural products from the NuBBE and UNPD databases. The regression line is dashed blackCC BY · PubMed Central
Fig. 4 SAscore and Tox-score prediction for several datasets. Violin plots show the distribution of (a) SAscore and (b) Tox-score values across active compounds from the Directory of Useful Decoys, Extended (DUD-E), FDA-approved drugs, natural products from the NuBBE and UNPD databases, and traditional Chinese medicines (TCM)
Fig. 4 SAscore and Tox-score prediction for several datasets. Violin plots show the distribution of (a) SAscore and (b) Tox-score values across active compounds from the Directory of Useful Decoys, Extended (DUD-E), FDA-approved drugs, natural products from the NuBBE and UNPD databases, and traditional Chinese medicines (TCM)CC BY · PubMed Central
Fig. 5 Performance of eToxPred in the prediction of toxic molecules. (a) The receiver operating characteristic plot and (b) the Matthews correlation coefficient (MCC) plotted as a function of the varying Tox-score. TPR and FPR are the true and false positive rates, respectively. Gray areas correspond to the performance of a random classifier. eToxPred is first applied to the primary training set (FDA-approved / TOXNET, solid black lines) to select the optimum Tox-score threshold. Then, the optimized eToxPred is applied to the independent testing set (KEGG-Drug and T3DB, solid black stars)
Fig. 5 Performance of eToxPred in the prediction of toxic molecules. (a) The receiver operating characteristic plot and (b) the Matthews correlation coefficient (MCC) plotted as a function of the varying Tox-score. TPR and FPR are the true and false positive rates, respectively. Gray areas correspond to the performance of a random classifier. eToxPred is first applied to the primary training set (FDA-approved / TOXNET, solid black lines) to select the optimum Tox-score threshold. Then, the optimized eToxPred is applied to the independent testing set (KEGG-Drug and T3DB, solid black stars)CC BY · PubMed Central
Fig. 6 Venn diagrams showing the overlap among various datasets. FDA-approved and TOXNET are the primary training datasets, whereas KEGG-Drug and T3DB are independent testing sets
Fig. 6 Venn diagrams showing the overlap among various datasets. FDA-approved and TOXNET are the primary training datasets, whereas KEGG-Drug and T3DB are independent testing setsCC BY · PubMed Central
Fig. 7 Performance of eToxPred in the prediction of specific toxicities. The receiver operating characteristic plots are shown for Carcinogenicity Potency (CP), cardiotoxicity (CD), endocrine disruption (ED), and acute oral toxicity (AO)
Fig. 7 Performance of eToxPred in the prediction of specific toxicities. The receiver operating characteristic plots are shown for Carcinogenicity Potency (CP), cardiotoxicity (CD), endocrine disruption (ED), and acute oral toxicity (AO)CC BY · PubMed Central
Fig. 8 Composition of non-toxic and toxic compounds. The scatter plot compares the frequencies of chemical fragments extracted with eMolFrag from FDA-approved (non-toxic) and TOXNET (toxic) molecules. The regression line is dotted black and the gray area delineates the corresponding confidence intervals. Three selected examples of fragments more commonly found in FDA-approved molecules (piperidine, piperazine, and fluorophenyl) are colored in green, whereas three counter examples of fragments more frequent in the TOXNET dataset (chlorophenyl, n-butyl, and acetic acid) are colored in red
Fig. 8 Composition of non-toxic and toxic compounds. The scatter plot compares the frequencies of chemical fragments extracted with eMolFrag from FDA-approved (non-toxic) and TOXNET (toxic) molecules. The regression line is dotted black and the gray area delineates the corresponding confidence intervals. Three selected examples of fragments more commonly found in FDA-approved molecules (piperidine, piperazine, and fluorophenyl) are colored in green, whereas three counter examples of fragments more frequent in the TOXNET dataset (chlorophenyl, n-butyl, and acetic acid) are colored in redCC BY · PubMed Central
Fig. 9 Composition of selected non-toxic compounds. Three examples of fragments more commonly found in FDA-approved molecules than in the TOXNET dataset: (a) piperidine, (b) piperazine, and (c) fluorophenyl. Four sample molecules containing a particular moiety (highlighted by green boxes) are selected from DrugBank and labeled by the DrugBank-ID
Fig. 9 Composition of selected non-toxic compounds. Three examples of fragments more commonly found in FDA-approved molecules than in the TOXNET dataset: (a) piperidine, (b) piperazine, and (c) fluorophenyl. Four sample molecules containing a particular moiety (highlighted by green boxes) are selected from DrugBank and labeled by the DrugBank-IDCC BY · PubMed Central
Fig. 10 Composition of selected toxic compounds. Three examples of fragments more commonly found in the TOXNET dataset than in FDA-approved molecules: (a) chlorophenyl, (b) n-butyl, and (c) acetic acid. Four sample molecules containing a particular moiety (highlighted by red boxes) are selected from ZINC and labeled by the ZINC-ID
Fig. 10 Composition of selected toxic compounds. Three examples of fragments more commonly found in the TOXNET dataset than in FDA-approved molecules: (a) chlorophenyl, (b) n-butyl, and (c) acetic acid. Four sample molecules containing a particular moiety (highlighted by red boxes) are selected from ZINC and labeled by the ZINC-IDCC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

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

The self-check needs the paper’s full text.

Citations

Cites

Show all 90 references

Research works printed but not linked: 2

  1. Theano_Development_Team: Theano: A Python framework for fast computation of mathematical expressions. arXiv e-prints 2016:abs/1605.02688.
  2. Tanimoto TT. An elementary mathematical theory of classification and prediction. In: Book An elementary mathematical theory of classification and prediction. (editor ed.^eds.). City; 1958.

Cited by

Sorted by citations

See all 212

Related

Metadata

Authors
Pu, L., Naderi, M., Liu, T., Wu, H.-C., Mukhopadhyay, S., & Brylinski, M.
Affiliations
Louisiana State University (US)
ORCID
Venue
BMC Pharmacology and Toxicology
Publisher
Springer Science and Business Media LLC
Published
2019-01-08
Volume
20
Issue
1
Pages
2-2
ISSN
2050-6511 (ISSN-L)
DOI
10.1186/s40360-018-0282-6
PMC
PMC6325674
OpenAlex
W2918544128
Cited by
212
References
90
Citations by year
2026: 21 · 2025: 33 · 2024: 50 · 2023: 44 · 2022: 29 · 2021: 20 · 2020: 10 · 2019: 5
FWCI
10.88
Type
journal article
Language
en
Open access
Licence
Locations
  • BMC Pharmacology and Toxicology (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Civil War Book Review (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
National Institute of General Medical Sciences (R35GM119524)
Topics
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › 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, drug candidates, synthetic accessibility, molecular fingerprints, machine learning, virtual screening, small organic molecules, drug-likeness, chemical toxicity
MeSH terms
Machine Learning, Algorithms, Animals, Humans, Drug Discovery, Drug-Related Side Effects and Adverse Reactions

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.

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