CoChat
Sign Up
Cited by 593Computational Drug Discovery MethodsMachine Learning in Materials ScienceProtein Degradation and Inhibitors

Exploiting machine learning for end-to-end drug discovery and development

Open access
Ekins, Sean, Puhl, Ana C., Zorn, Kimberley M., Lane, Thomas R., Russo, Daniel P., Klein, Jennifer J., Hickey, Anthony J., Clark, Alex M.·2019·Nature Materials
DOI 10.1038/s41563-019-0338-zPMC PMC6594828
Open this paper in CoChat — free account

Abstract

A variety of machine learning methods such as naive Bayesian, support vector machines and more recently deep neural networks are demonstrating their utility for drug discovery and development. These leverage the generally bigger datasets created from high-throughput screening data and allow prediction of bioactivities for targets and molecular properties with increased levels of accuracy. We have only just begun to exploit the potential of these techniques but they may already be fundamentally changing the research process for identifying new molecules and/or repurposing old drugs. The integrated application of such machine learning models for end-to-end (E2E) application is broadly relevant and has considerable implications for developing future therapies and their targeting.This Perspective describes the application of machine learning models in the design, synthesis and characterisation of molecules at different stages in the drug discovery and development process.

AI snapshot

Written from the paper’s full text.

TL;DR

This review argues that machine learning can be integrated end-to-end across drug discovery and development, leveraging large public datasets and models for bioactivity, ADME/Tox, repurposing, and nanomedicine. It highlights that deep neural networks often excel in cross-validation but do not consistently outperform other methods on external testing, while data readiness, accessibility, and interpretability remain barriers.

Key findings

  • The paper proposes an end-to-end (E2E) vision in which machine learning models span the R&D pathway to predict clinical viability, but notes that linear combinations of models may accumulate errors depending on each model’s accuracy.
  • In cited comparisons, DNNs ranked higher than all other machine learning methods across all datasets after 4-fold cross validation using FCFP6 fingerprints; for 18,886 compounds screened against *Mycobacterium tuberculosis*, DNNs and support vector machines appeared superior for training and 5-fold cross validation, but external testing of DNN models with a large test set did not perform as well as other methods. For estrogen receptor datasets, DNNs had higher accuracy within the training set in 5-fold cross validation, while external test predictions were similar between DNN and most classic machine learning models.
  • Public chemical and biological data in PubChem, ChEMBL, and other databases now include millions of molecules and bioactivities; tuberculosis and malaria have high-throughput screening datasets of >300,000 compounds, and rare diseases are defined as affecting less than 200,000 people in the United States.
  • Applications cited include repurposing topiramate for inflammatory bowel disease, imipramine, bepridil, promethazine, and cimetidine as non-small cell lung cancer inhibitors, and CDRscan identifying 14 oncology and 23 non-oncology drugs with new potential cancer indications; a random forest classifier identified category C drugs likely harmful or safe for fetal loss or congenital anomalies.

Methods

  • Design: This is a review/perspective article, not a primary experimental study; it synthesizes literature and the authors’ prior cheminformatics efforts.
  • Data or sample: It cites public databases such as PubChem and ChEMBL, high-throughput screening datasets, and prior datasets including 18,886 *M. tuberculosis* compounds, estrogen receptor datasets, and >300,000 compounds for tuberculosis and malaria.
  • Main analysis: Narrative synthesis and discussion of machine learning algorithms, molecular descriptors, validation strategies, and drug discovery applications; no systematic review or meta-analysis methodology is stated.

Limitations

  • The paper states that many databases are not “model-ready” or machine-readable and require expert curation; DNNs are not “plug and play” and take significant time to optimize; there is no agreement on the best algorithm; DNNs have rarely been used for prospective prediction; generative DNN proof-of-concept studies often have not synthesized molecules to validate predictions; nanoparticle datasets are small (10–100s of molecules), limiting deep learning; and linear model combinations may accumulate errors.
  • Evident from the text: as a review, it reports no new primary data or prospective validation of its own, and its conclusions depend on the cited studies rather than a systematic search or quantitative synthesis.

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

Integrity signals

Registry

Crossref · 10.1038/s41563-019-0338-z

Cites

95 of 95 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 89 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Our own recent cheminformatics prospective testing efforts have identified compounds active in vitro and in vivo against Chagas disease and the Ebola virus using Bayesian algorithms.

    “Several of our own recent cheminformatics prospective testing efforts have identified compounds active in vitro and in vivo against Chagas disease13 and the Ebola virus41 using Bayesian algorithms.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [13] Ekins, S. et al. Machine learning models and pathway genome data base for trypanosoma cruzi drug discovery. PLoS Negl. Trop. Dis. 9, e0003878 (2015). source

      Not checked in the free run.

    • cites [41] Ekins, S. et al. Machine learning models identify molecules active against Ebola virus in vitro. F1000Research 4, 1091 (2015). source

      Not checked in the free run.

  • Findingcitation

    After 4-fold cross validation and ranked normalized scores of metrics, DNNs ranked higher than all the other machine learning methods across all datasets.

    “After 4-fold cross validation and ranked normalized scores of metrics, DNNs ranked higher than all the other machine learning methods across all datasets48.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [48] Stratton, T. P. et al. Addressing the metabolic stability of antituberculars through machine learning. ACS Med. Chem. Lett. 8, 1099–1104 (2017). source

      Not checked in the free run.

  • Findingcitation

    For external test set predictions on multiple estrogen receptor datasets, DNN and most classic machine learning models perform similarly regardless of dataset or molecular descriptors.

    “For external test set predictions DNN and most classic machine learning models perform similarly regardless of dataset or molecular descriptors46.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [46] Russo, D. P., Zorn, K. M., Clark, A. M., Zhu, H. & Ekins, S. Comparing multiple machine learning algorithms and metrics for estrogen receptor binding prediction. Mol. Pharm. 15, 4361–4370 (2018). source

      Not checked in the free run.

  • Findingcitation

    For predicting compounds within the training set for multiple estrogen receptor datasets, DNNs had higher accuracy than other methods in 5-fold cross validation.

    “For predicting compounds within the training set, DNNs had higher accuracy than other methods in 5-fold cross validation.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [46] Russo, D. P., Zorn, K. M., Clark, A. M., Zhu, H. & Ekins, S. Comparing multiple machine learning algorithms and metrics for estrogen receptor binding prediction. Mol. Pharm. 15, 4361–4370 (2018). source

      Not checked in the free run.

  • Findingown data

    External testing of DNN models with a large test set did not perform as well as other machine learning methods.

    “Conversely, external testing of DNN models with a large test set did not perform as well as other machine learning methods.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [51] Koutsoukas, A., Monaghan, K. J., Li, X. & Huan, J. Deep-learning: investigating deep neural networks hyper-parameters and comparison of performance to shallow methods for modeling bioactivity data. J. Cheminform. 9, 42 (2017). source

      Not checked in the free run.

  • May combine two claims. 33
  • Classified as probably not a substantial claim. 36

  • May combine two claims.

    Findingcitation

    We have used many different machine learning algorithms and descriptors in parallel to identify the optimum combination and address complex problems related to improving solubility or metabolic stability while retaining bioactivity.

    “We have also used many different machine learning algorithms and descriptors in parallel to identify the optimum combination46 and address complex problems facing the pharmaceutical industry related to the challenges of improving solubility or metabolic stability47 while retaining bioactivity.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [46] Russo, D. P., Zorn, K. M., Clark, A. M., Zhu, H. & Ekins, S. Comparing multiple machine learning algorithms and metrics for estrogen receptor binding prediction. Mol. Pharm. 15, 4361–4370 (2018). source

      Not checked in the free run.

    • cites [47] Lusci, A., Pollastri, G. & Baldi, P. Deep architectures and deep learning in chemoinformatics: the prediction of aqueous solubility for drug-like molecules. J. Chem. Inf. Model. 53, 1563–1575 (2013). source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    Our own efforts continue to reflect the pattern that while DNN excels at cross validation assessments it is generally no better than other machine learning methods for external testing.

    “Our own efforts continue to reflect this pattern, namely that while DNN excels at cross validation assessments it is generally no better than other machine learning methods for external testing.”
    Quote in the paperSubstantial claim

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

    Interpretationnone given

    Machine learning methods leverage the generally bigger data sets created from high throughput screening data and allow prediction of bioactivities for targets and molecular properties with increased levels of accuracy.

    “These leverage the generally bigger data sets created from high throughput screening data and allow prediction of bioactivities for targets and molecular properties with increased levels of accuracy.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Less than 20 years ago, developing a drug took ~12 years, cost under a billion dollars, and the biggest challenges were failures due to efficacy or toxicity-induced attrition.

    “Less than 20 years ago, developing a drug took ~12 years, cost under a billion dollars, and the biggest challenges were failures due to efficacy or toxicity-induced attrition2.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [2] Kola, I. & Landis, J. Can the pharmaceutical industry reduce attrition rates. Nat. Rev. Drug. Discov. 3, 711–715 (2004). source

      Not checked in the free run.

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

    Premisecitation

    2D structures of molecules have been used as an input to predict toxicity using the Tox21 benchmark set, which is also part of a platform called MoleculeNet.

    “2D structures of molecules have seen use as an input to predict toxicity using the Tox21 benchmark set34, which is also part of a platform called MoleculeNet35.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [34] Butler, K. T., Davies, D. W., Cartwright, H., Isayev, O. & Walsh, A. Machine learning for molecular and materials science. Nature 559, 547–555 (2018). source

      Not checked in the free run.

    • cites [35] Rifaioglu, A. S. et al. Recent applications of deep learning and machine intelligence on in silico drug discovery: methods, tools and databases. Brief Bioinform. https://doi.org/10.1093/bib/bby061 (2018). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Machine learning will impact the future of design, synthesis, characterization and application of molecules and materials.

    “These efforts build on the recent proposal that machine learning will impact the future of design, synthesis, characterization and application of molecules and materials10.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10] Davies, D. W., Butler, K. T., Isayev, O. & Walsh, A. Materials discovery by chemical analogy: role of oxidation states in structure prediction. Faraday Discuss. 211, 553–568 (2018). source

      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

    Expert curation is always required to ensure compatibility of data when downloading data in electronic formats, and domain expertise is important to build and use machine learning models.

    “When you can download data in electronic formats, expert curation is always required to ensure compatibility of data and domain expertise is important to build and use machine learning models.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    DNNs have been used in pattern recognition and machine learning, sparking their use in pharmacology and drug discovery and becoming a source for numerous recent reviews.

    “DNNs have been used in pattern recognition and machine learning25, sparking their use in pharmacology and drug discovery26 and becoming a source for numerous recent reviews12.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [25] Schmidhuber, J. Deep learning in neural networks: an overview. Neural Netw. 61, 85–117 (2015). source

      Not checked in the free run.

    • cites [26] Aliper, A. et al. Deep learning applications for predicting pharmacological properties of drugs and drug repurposing using transcriptomic data. Mol. Pharm. 13, 2524–2530 (2016). source

      Not checked in the free run.

    • cites [12] Chen, H., Engkvist, O., Wang, Y., Olivecrona, M. & Blaschke, T. The rise of deep learning in drug discovery. Drug Discov. Today. 23, 1241–1250 (2018). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation

    DNNs have rarely been used for prospective prediction, and this is a limitation of many machine learning drug discovery studies published.

    “but to date this algorithm has rarely been used for prospective prediction and in some respects this is a limitation of many of the machine learning drug discovery studies published.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [32] Altae-Tran, H., Ramsundar, B., Pappu, A. S. & Pande, V. Low data drug discovery with one-shot learning. ACS Cent. Sci. 3, 283–293 (2017). source

      Not checked in the free run.

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

    Premisecitation

    Computational phenotyping with machine learning has been used to predict antibiotic resistance phenotypes in a variety of bacterial species.

    “an approach called computational phenotyping has emerged to embrace the complexity inherent in disease mechanisms with machine learning to define accurate phenotypes and has been used to predict antibiotic resistance phenotypes in a variety of bacterial species11.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [11] Drouin, A. et al. Predictive computational phenotyping and biomarker discovery using reference-free genome comparisons. BMC Genom. 17, 754 (2016). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation

    For many targets and diseases we now have plentiful data, as indicated by the wide use of ChEMBL and the accuracy of the models generated.

    “We would suggest for many of the targets and diseases we now have plentiful data as indicated by the wide use of ChEMBL and the accuracy of the models generated18.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [18] Mayr, A. et al. Large-scale comparison of machine learning methods for drug target prediction on ChEMBL. Chem. Sci. 9, 5441–5451 (2018). source

      Not checked in the free run.

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

    Premisecitation

    DNN algorithms are increasingly available, but they are not 'plug and play' and their use takes significant time to optimize.

    “While DNN algorithms are increasingly available9, they are not ‘plug and play’ and their use takes significant time to optimize.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] Zhavoronkov, A. Artificial intelligence for drug discovery, biomarker development, and generation of novel chemistry. Mol. Pharm. 15, 4311–4313 (2018). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationcitation

    Implementing machine learning across all areas of drug discovery and development for which there is sufficient data to train models is another way to possibly improve productivity in this complex environment.

    “Another way to possibly improve productivity in this complex environment is to implement machine learning across all areas of drug discovery and development9 for which there is sufficient data to train models.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] Zhavoronkov, A. Artificial intelligence for drug discovery, biomarker development, and generation of novel chemistry. Mol. Pharm. 15, 4311–4313 (2018). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Models for multiple targets already assist in the prediction of off-target effects as well as predicting the most potent compound-target interactions.

    “Models for multiple targets already assist in the prediction of off-target effects14 as well as predicting the most potent compound-target interactions15.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [14] Lampa, S. et al. Predicting off-target binding profiles with confidence using conformal prediction. Front. Pharmacol. 9, 1256 (2018). source

      Not checked in the free run.

    • cites [15] Reker, D., Rodrigues, T., Schneider, P. & Schneider, G. Identifying the macromolecular targets of de novo-designed chemical entities through self-organizing map consensus. Proc. Natl Acad. Sci. USA 111, 4067–4072 (2014). source

      Not checked in the free run.

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

    Premisecitation

    The use of DNNs is still in its relative infancy and has limited applications for cheminformatics as compared with other methods.

    “The use of DNNs is still in its relative infancy and has limited applications for cheminformatics as compared with other methods30.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [30] Ekins, S. The next era: Deep learning in pharmaceutical research. Pharm. Res. 33, 2594–2603 (2016). source

      Not checked in the free run.

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

    Interpretationnone given

    Potential limitations of a linear combination of models might appear as errors could accumulate depending on the accuracy of each model, which may then influence the overall utility and prediction.

    “Potential limitations of a linear combination of models might appear as errors could accumulate depending on the accuracy of each model, which may then influence the overall utility and prediction.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    DNNs have been used in various pharmaceutical applications from docking to virtual screening and beyond.

    “DNNs have been used in various pharmaceutical applications from docking to virtual screening and beyond (Table 1), but the rise in prominence is linked to increased computational power and the availability of larger datasets.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Much of the interest in DNNs is centered around the flexibility of their architecture, which allows the generation of models for single task or multitask machine learning as well as predicting drug-target interactions.

    “much of the interest in them is centered around the flexibility of their architecture27 which allows the generation of models for single task or multitask machine learning28 as well as predicting drug-target interactions29.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [27] Ma, J., Sheridan, R. P., Liaw, A., Dahl, G. E. & Svetnik, V. Deep neural nets as a method for quantitative structure-activity relationships. J. Chem. Inf. Model. 55, 263–274 (2015). source

      Not checked in the free run.

    • cites [28] Wu, K., Zhao, Z., Wang, R. & Wei, G.-W. TopP–S: Persistent homology-based multi-task deep neural networks for simultaneous predictions of partition coefficient and aqueous solubility. J. Comput. Chem. 39, 1444–1454 (2018). source

      Not checked in the free run.

    • cites [29] Wen, M. et al. Deep-learning-based drug-target interaction prediction. J. Proteome Res. 16, 1401–1409 (2017). source

      Not checked in the free run.

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

    Interpretationnone given

    The rise in prominence of DNNs is linked to increased computational power and the availability of larger datasets.

    “but the rise in prominence is linked to increased computational power and the availability of larger datasets.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationreasoning

    Machine learning can be applied for the same cost to identify treatments for a variety of diseases and may level the drug development field for smaller companies and researchers.

    “The impact of this viewpoint is that machine learning can be applied for the same cost to identify treatments for a variety of diseases and may level the drug development field for smaller companies and researchers.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Recent interest in deep learning or deep neural networks (DNNs) for drug discovery has catalyzed interest in machine learning in this field more broadly.

    “However, recent interest in deep learning or deep neural networks (DNNs) for drug discovery has catalyzed interest in machine learning in this field more broadly.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Machine learning models such as support vector machines, k-Nearest Neighbors, Naïve Bayes, Random Forest and many other methods have long been utilized for drug discovery.

    “Machine learning models such as support vector machines20, k-Nearest Neighbors21, Naïve Bayes22, Random Forest23 and many other methods24, have long been utilized for drug discovery.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [20] Christianini, N. & Shawe-Taylor, J. Support Vector Machines and Other Kernel-Based Learning Methods (Cambridge Univ. Press, 2000). source

      Not checked in the free run.

    • cites [21] Shen, M., Xiao, Y., Golbraikh, A., Gombar, V. K. & Tropsha, A. Development and validation of K-nearest neighbour QSPR models of metabolic stability of drug candidates. J. Med. Chem. 46, 3013–3020 (2003). source

      Not checked in the free run.

    • cites [22] Bender, A. et al. Analysis of pharmacology data and the prediction of adverse drug reactions and off-target effects from chemical structure. ChemMedChem 2, 861–873 (2007). source

      Not checked in the free run.

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

    Interpretationreasoning

    These challenges still persist partly because the datasets may not cover sufficient chemical space, and the test molecules could be outside the applicability of the training set of the models.

    “These challenges still persist partly because the datasets may not cover sufficient chemical space, and the test molecules could be outside the applicability of the training set of the models.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Machine learning methods in the pharmaceutical industry are most commonly used for virtual screening of compounds, reducing the need to generate more high-throughput screening data by cherry-picking compounds and performing low to medium-throughput screening.

    “Machine learning methods in the pharmaceutical industry are most commonly used for virtual screening of compounds, reducing the need to generate more high-throughput screening data by cherry-picking compounds and performing low to medium-throughput screening38.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [38] McGaughey, G. B. et al. Comparison of topological, shape, and docking methods in virtual screening. J. Chem. Inf. Model. 47, 1504–1519 (2007). source

      Not checked in the free run.

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

    Premisecitation

    The same machine learning algorithms have been used widely in both pharmaceutical and toxicological research.

    “The same machine learning algorithms have been used widely in both pharmaceutical and toxicological research30 (Table 1).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [30] Ekins, S. The next era: Deep learning in pharmaceutical research. Pharm. Res. 33, 2594–2603 (2016). source

      Not checked in the free run.

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

    Interpretationnone given

    The requirement for expert users has been the Achilles' heel of cheminformatics, whereas computational tools for bioinformatics have found broader use due to their accessibility.

    “The requirement for expert users in many ways has been the Achilles’ heel of cheminformatics, whereas computational tools for bioinformatics have found broader use due to their accessibility.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Generative DNNs have been described for the generation of virtual libraries of molecules and enable de novo drug design with optimized properties.

    “Generative DNNs have also been described for the generation virtual libraries of molecules and these enable de novo drug design with optimized properties36.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [36] Popova, M., Isayev, O. & Tropsha, A. Deep reinforcement learning for de novo drug design. Sci. Adv. 4, eaap7885 (2018). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationreasoning

    We are at a clear tipping point for machine learning and deep learning in particular, but full integration of these models is still likely to be a work in progress.

    “We are at a clear tipping point for machine learning and deep learning in particular, but it has taken decades to reach this point and yet full integration of these models is still likely to be a work in progress.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    There are neglected and tropical diseases and rare diseases in which interest has increased as the FDA priority voucher has provided an incentive for companies to develop new treatments.

    “There are neglected and tropical diseases (e.g. malaria, tuberculosis and others) and rare diseases (defined as affecting less than 200,000 people in the United States) in which interest has increased as the FDA priority voucher52 has provided an incentive for companies to develop new treatments.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [52] Lane, T. et al. Comparing and validating machine learning models for mycobacterium tuberculosis drug discovery. Mol. Pharm. 15, 4346–4360 (2018). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    Several machine learning frameworks and tools available today aimed at using small molecules and related data are not at the point where they are universally accessible to all scientists.

    “While there are several machine learning frameworks and tools available today aimed at using small molecules and related data, they are therefore not at the point where they are universally accessible to all scientists.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    While NIH funding has gone into high throughput screening, until recently there have not been comparable efforts on data mining or machine learning with this data.

    “While NIH funding has gone into the high throughput screening54, until recently there have not been comparable efforts on data mining or machine learning with this data16.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [54] Ekins, S. et al. Bayesian models leveraging bioactivity and cytotoxicity information for drug discovery. Chem. Biol. 20, 370–378 (2013). source

      Not checked in the free run.

    • cites [16] Kim, S. et al. PubChem substance and compound databases. Nucleic Acids Res. 44, D1202–1213 (2016). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Machine learning methods have been applied to electronic health records to accurately predict multiple medical events from different centers without site-specific data harmonization, with recent data suggesting that deep learning was comparable to regularized logistic regression in this case.

    “For example, machine learning methods have been applied to electronic health records to accurately predict multiple medical events from different centers without site-specific data harmonization, with recent data suggesting that deep learning was comparable to regularized logistic regression in this case40.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [40] Rajkomar, A. et al. Scalable and accurate deep learning with electronic health records. npj Digit. Med. 1, 18 (2018). source

      Not checked in the free run.

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

    Interpretationcitation

    Machine learning could be used to find treatments for rare diseases using an iterative approach.

    “We have also proposed that machine learning could be used to find treatments for rare diseases using an iterative approach56.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [56] Ekins, S. Industrializing rare disease therapy discovery and development. Nat. Biotechnol. 35, 117–118 (2017). source

      Not checked in the free run.

    • Rests on an assumption: To date there are no efforts specifically using machine learning to identify drugs for rare diseases that leverage the relevant datasets for targets that are in the public domain.

  • May combine two claims.

    Interpretationcitation

    DNN does not always outperform other methods for external testing and in several cases is not the best, which is important to consider due to the computational cost of DNN.

    “The fact that DNN does not always outperform other methods for external testing and in several cases is not the best, is important to consider due to the computational cost of DNN.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    Others have initiated different approaches to combat rare diseases through the development of a comprehensive global genotype-phenotype database, sharing genomics data or other aspects of rare diseases, as well as assist patients and caregivers.

    “Others have initiated different approaches to combat rare diseases through the development of a comprehensive global genotype-phenotype database58, sharing genomics data59 or other aspects of rare diseases60, as well as assist patients and caregivers61.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [58] Chen, B. & Altman, R. B. Opportunities for developing therapies for rare genetic diseases: focus on gain-of-function and allostery. Orphanet. J. Rare Dis. 12, 61 (2017). source

      Not checked in the free run.

    • cites [59] Trujillano, D. et al. A comprehensive global genotype-phenotype database for rare diseases. Mol. Genet. Genomic Med. 5, 66–75 (2017). source

      Not checked in the free run.

    • cites [60] Thompson, R. et al. RD-Connect: an integrated platform connecting databases, registries, biobanks and clinical bioinformatics for rare disease research. J. Gen. Intern. Med. 29, 780–787 (2014). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    The closest example to this ideal scenario purchased close analogs to the molecules generated with generative DNNs and tested them against different kinases at 10μM, identifying several hits.

    “The closest example to this ideal scenario has purchased close analogs to the molecules generated with generative DNNs and tested them against different kinases at 10μM, identifying several hits37.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [37] Putin, E. et al. Adversarial threshold neural computer for molecular de novo design. Mol. Pharm. 15, 4386–4397 (2018). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Other researchers have compared several machine learning approaches with different datasets from ChEMBL using random split and temporal cross validation to show the superiority of DNNs, or 5-fold cross validation and leave out 40% as a validation set.

    “Other researchers have also compared several machine learning approaches with different datasets from ChEMBL using random split and temporal cross validation to show the superiority of DNNs49, or 5-fold cross validation and leave out 40% as a validation set50.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [49] Korotcov, A., Tkachenko, V., Russo, D. P. & Ekins, S. Comparison of deep learning with multiple machine learning methods and metrics using diverse drug discovery datasets. Mol. Pharm. 14, 4462–4475 (2018). source

      Not checked in the free run.

    • cites [50] Lenselink, E. B. et al. Beyond the hype: deep neural networks outperform established methods using a ChEMBL bioactivity benchmark set. J. Cheminform. 9, 45 (2017). source

      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

    There is now increasing focus on machine learning in drug discovery which suggests the utility of such approaches is increasing.

    “There is now increasing focus on machine learning in drug discovery which suggests the utility of such approaches is increasing.”
    Quote in the paperNot a substantial claim

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

    Interpretationcitation

    These approaches could learn from the work performed on neglected and tropical diseases which has used public datasets to identify new compounds.

    “These approaches could learn from the work performed on neglected and tropical diseases which has used public datasets to identify new compounds54.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [54] Ekins, S. et al. Bayesian models leveraging bioactivity and cytotoxicity information for drug discovery. Chem. Biol. 20, 370–378 (2013). source

      Not checked in the free run.

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

    Premisecitation

    The number of deals between start-up machine learning-based drug discovery companies and big pharma, biotechs, or VC investors exemplifies the increasing utility of such approaches.

    “This is exemplified by the number of deals between start-up machine learning-based drug discovery companies and big pharma, biotechs, or VC investors62.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [62] Rare Disease InfoHub https://rarediseases.oscar.ncsu.edu (2018).

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    In inflammatory bowel disease, topiramate was identified by intersecting disease signatures with Connectivity Map data using a pattern-matching algorithm and experimentally validated as a novel treatment.

    “For example, in a classic paradigm inflammatory bowel disease (IBD) signatures were derived from surgical specimens and intersected with Connectivity Map70 data representing transcriptional readouts across a number of cell lines in response to treatment with many hundreds of drugs using a novel pattern-matching algorithm71. From this research the anticonvulsant drug topiramate was identified and”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [70] Baker, N. C., Ekins, S., Williams, A. J. & Tropsha, A. A bibliometric review of drug repurposing. Drug Discov. Today 23, 661–672 (2018). source

      Not checked in the free run.

    • cites [71] Lamb, J. et al. The connectivity map: using gene-expression signatures to connect small molecules, genes, and disease. Science 313, 1929–1935 (2006). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    As most of the published examples use small datasets (10-100s of molecules), deep learning has limited value and has rarely been applied in nanoparticle research.

    “As most of the published examples use small datasets (10–100s of molecules), deep learning has limited value and has rarely been applied.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    If we can combine high-quality curated screening data with cutting edge machine learning algorithms and molecular descriptors, there is the opportunity to build models that can be used to reliably predict new molecules for most areas of drug discovery.

    “If we can combine high-quality curated screening data with cutting edge machine learning algorithms and molecular descriptors, there is the opportunity to build models that can be used to reliably predict new molecules for most areas of drug discovery.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    Using machine learning more broadly across the industry could allow moving beyond the limitations defined by researcher specialty and data silos.

    “Using machine learning more broadly across the industry could allow us to move beyond the limitations defined by researcher specialty and data silos but it will be important to perform prospective validation of the models to demonstrate progress.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    Applying machine learning to the whole pathway can evaluate how a network of protein interactions will react to a perturbation such as a drug acting on a particular target, and this could lead to more personalized medicine.

    “In this context, we can apply machine learning to the whole pathway to evaluate how a network of protein interactions will react to a perturbation in the system such as the drug that is acting on a particular target and this in turn could lead to more personalized medicine75.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [75] Cruz, S. et al. In silico HCT116 human colon cancer cell-based models en route to the discovery of lead-like anticancer drugs. Biomolecules 8, 56 (2018). source

      Not checked in the free run.

    • Rests on an assumption: Proteins have isoforms and redundancy, so inhibition of one protein might not be enough to elicit the desired response.
    • Rests on an assumption: Inhibition of one pathway might not be enough to achieve the response because cells have other pathway mechanisms that would be activated to circumvent the inhibited pathway.

  • May combine two claims.

    Interpretationreasoning

    These efforts can enable the quantitative prediction of desirable molecules before synthesis and focus research on experiments with the most promising candidates.

    “These efforts can enable the quantitative prediction of desirable molecules before synthesis and focus research on experiments with the most promising candidates.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Machine learning can help explore the patterns of known drugs and their interactions with drug targets and potentially repurpose already approved molecules.

    “Our thinking need not be limited to discovering just new molecular entities as machine learning can help explore the patterns of known drugs and their interactions with drug targets and potentially repurpose already approved molecules.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    Machine learning methods have been applied to predict the off-targets of CRISPR-Cas9 gene editing.

    “CRISPR-Cas9, has the potential to edit and renovate the harmful genes for personalized therapy and machine learning methods have been applied to predict the off-targets of CRISPR-Cas9 gene editing77.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [77] Chen, R., Liu, X., Jin, S., Lin, J. & Liu, J. Machine learning for drug-target interaction prediction. Molecules 23, 2208 (2018). source

      Not checked in the free run.

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

    Interpretationreasoning

    Limitations of computational models due to the vastness of chemical space and the potential for selecting compounds with unfavorable properties would be mitigated to some extent if previously approved drugs with extensive ADMET data, or better still phase I data, could be repurposed.

    “Often a major concern about making predictions with computational models is the vastness of chemical space and the potential for selecting compounds with unfavorable properties. These limitations would be mitigated to some extent if previously approved drugs that have extensive ADMET data (or better still have reached phase I trials) could be repurposed.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: A major concern about making predictions with computational models is the vastness of chemical space and the potential for selecting compounds with unfavorable properties.

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

    Premisecitation

    We and others have taken different routes to improve interpretability of machine learning models, including tools to highlight contributions of models to test molecules, identifying training compounds in the same neighborhood as test molecules and scores of model applicability or overlap.

    “We and others have taken different routes to improve this aspect including tools to highlight contributions of models to test molecules64, identifying training compounds in the same neighborhood as test molecules and scores of model applicability or overlap46, 63.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [64] Chuang, K. V. & Keiser, M. J. Adversarial controls for scientific machine learning. ACS Chem. Biol. 13, 2819–2821 (2018). source

      Not checked in the free run.

    • cites [46] Russo, D. P., Zorn, K. M., Clark, A. M., Zhu, H. & Ekins, S. Comparing multiple machine learning algorithms and metrics for estrogen receptor binding prediction. Mol. Pharm. 15, 4361–4370 (2018). source

      Not checked in the free run.

    • cites [63] Fleming, N. How artificial intelligence is changing drug discovery. Nature 557, 55–57 (2018). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Using machine learning to integrate diverse, large-scale data leads to selecting the best targets and may ultimately help predict efficacy.

    “This leads us to selecting the best targets and may ultimately help us to predict efficacy.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    Some approaches using machine learning methods have been developed to detect drug-target interactions, which is fundamental to both new drug discovery and drug repositioning.

    “Some approaches using machine learning methods have been developed to detect drug-target interactions76, which is fundamental to both new drug discovery and drug repositioning.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [76] Fröhlich, H. et al. From hype to reality: data science enabling personalized medicine. BMC Med. 16, 150 (2018). source

      Not checked in the free run.

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

    Premisenone given

    The field of nanomedicine has led to the development of nanoinformatics and the use of data mining and machine learning to develop nano-QSARs to predict functional and structural properties of nanoparticles.

    “The field of nanomedicine has led to the development of nanoinformatics and the use of data mining and machine learning to develop nano-QSARs to predict functional and structural properties of nanoparticles.”
    Quote in the paperNot a substantial claim

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

    Interpretationcitation

    There are likely many examples of approved drugs finding additional uses for various diseases.

    “There are likely many examples of approved drugs finding additional uses for various diseases54”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [54] Ekins, S. et al. Bayesian models leveraging bioactivity and cytotoxicity information for drug discovery. Chem. Biol. 20, 370–378 (2013). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Quantitative structure-nanoparticle assembly prediction models have been used to generate predictions of nano-assembly which were found to encapsulate drugs with high loadings and have then also been validated in cancer models.

    “Quantitative structure-nanoparticle assembly prediction models have been used to generate predictions of nano-assembly which were found to encapsulate drugs with high loadings and have then also been validated in cancer models66.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [66] Jones, D. E., Ghandehari, H. & Facelli, J. C. A review of the applications of data mining and machine learning for the prediction of biomedical properties of nanoparticles. Comput. Methods Programs Biomed. 132, 93–103 (2016). source

      Not checked in the free run.

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

    Premisecitation

    A machine learning method has been described to identify clinical trials involving nanodrugs and nanodevices from ClinicalTrials.gov.

    “Interestingly, a machine learning method has also been described to identify clinical trials involving nanodrugs and nanodevices from ClinicalTrials.gov67.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [67] Shamay, Y. et al. Quantitative self-assembly prediction yields targeted nanomedicines. Nat. Mater. 17, 361–368 (2018). source

      Not checked in the free run.

56 of 56

  • The paper relies on this without support.

    The biggest cost and most time-consuming component associated with drug development is conducting clinical trials.

    “The biggest cost and most time-consuming component associated with drug development is conducting clinical trials.”

  • The paper relies on this without support.

    Currently available chemical and biological data relevant for rare disease drug discovery is available but diffuse, existing in an array of public or private databases.

    “Currently available chemical and biological data relevant for rare disease drug discovery is available but diffuse, existing in an array of public or private databases.”

  • The paper relies on this without support.

    To date there are no efforts specifically using machine learning to identify drugs for rare diseases that leverage the relevant datasets for targets that are in the public domain.

    “To date there are no efforts specifically using machine learning to identify drugs for rare diseases that leverage the relevant datasets for targets that are in the public domain.”

  • The paper relies on this without support.

    Commercially available tools that could enable scientists to develop nanomedicines using these models have yet to be developed to date.

    “Commercially available tools that could enable scientists to develop nanomedicines using these models have yet to be developed to date.”

  • The paper relies on this without support.

    Proteins have isoforms and redundancy, so inhibition of one protein might not be enough to elicit the desired response.

    “Proteins have isoforms and redundancy, thus, inhibition of one might not be enough to illicit the desired response.”

  • The paper relies on this without support.

    Inhibition of one pathway might not be enough to achieve the response because cells have other pathway mechanisms that would be activated to circumvent the inhibited pathway.

    “In the same way, inhibition of one pathway might not be enough to achieve the response, since the cell has other pathway mechanisms that would be activated to circumvent the one that has been inhibited.”

  • The paper relies on this without support.

    There is currently an abundance of data, powerful and plentiful computing, public and private efforts to develop databases and models, and accessible fee-for-service ways to test compounds.

    “We now have an abundance of data, powerful and plentiful computing, public and private efforts to develop databases and models and accessible ways to test compounds on a fee-for-service basis.”

  • The paper relies on this without support.

    A major concern about making predictions with computational models is the vastness of chemical space and the potential for selecting compounds with unfavorable properties.

    “Often a major concern about making predictions with computational models is the vastness of chemical space and the potential for selecting compounds with unfavorable properties.”
8 of 8

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?

  • Mining of large-scale chemistry data is needed to efficiently classify potential drug compounds, and machine learning techniques have shown great potential.

    “However, mining of the large-scale chemistry data is needed to efficiently classify potential drug compounds and machine learning techniques have shown great potential [15].”

    From The rise of artificial intelligence in healthcare applications (2020) · cites this paper as [15]

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.

  • The same reference has two publication years

    Minor

    References 48 and 83 list the same authors, journal, volume, pages, and article identifiers, but give its publication year as 2018 and 2017, respectively.

1 of 1

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. Korotcov, A., Tkachenko, V., Russo, D. P. & Ekins, S. Comparison of deep learning with multiple machine learning methods and metrics using diverse drug discovery data sets. Mol. Pharm. 14, 4462–4475 (2017). · DOI

Cited by

Sorted by citations

Related

Authors
Ekins, S., Puhl, A. C., Zorn, K. M., Lane, T. R., Russo, D. P., Klein, J. J., Hickey, A. J., & Clark, A. M.
Affiliations
  • Collaborations Pharmaceuticals (United States) (US)
  • Rutgers, The State University of New Jersey (US)
  • University of North Carolina at Chapel Hill (US)
Corresponding author
Ekins, Sean
ORCID
Venue
Nature Materials
Publisher
Springer Science and Business Media LLC
Published
2019-04-18
Volume
18
Issue
5
Pages
435-441
ISSN
  • 1476-1122 (ISSN-L)
  • 1476-4660
DOI
10.1038/s41563-019-0338-z
PMC
PMC6594828
OpenAlex
W2940242941
Semantic Scholar
daf468f001c3a5c6f9e667417becb94fa83efb2f
Cited by
578
References
95
Citations by year
2026: 79 · 2025: 85 · 2024: 85 · 2023: 63 · 2022: 80 · 2021: 95 · 2020: 72 · 2019: 19
FWCI
45.24
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Materials (publishedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (acceptedVersion)
  • Carolina Digital Repository (University of North Carolina at Chapel Hill) (submittedVersion)
Funders
  • National Institutes of Health (R44TR000942, UH2TR002084)
  • National Institute of General Medical Sciences (R44GM122196)
  • National Institute of Allergy and Infectious Diseases (U19AI098461)
  • National Institute of Neurological Disorders and Stroke (R43NS107079)
  • National Center for Advancing Translational Sciences
Topics
  • Physical Sciences › Computer Science › Computational Theory and Mathematics › Computational Drug Discovery Methods
  • Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Degradation and Inhibitors
Keywords
machine learning, drug discovery, drug development, high-throughput screening data, bioactivity prediction, molecular property prediction, drug repurposing, deep neural network, support vector machine, naive Bayesian classifier, molecular design
MeSH terms
Machine Learning, Drug Development, Algorithms, Bayes Theorem, Computer Simulation, Humans, Technology, Pharmaceutical, Drug Design, Neural Networks, Computer, Computational Biology, Nanomedicine, Drug Discovery, Drug Repositioning, Support Vector Machine

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 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

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