CoChat
Sign Up
Cited by 146Explainable Artificial Intelligence (XAI)Computational Drug Discovery MethodsMachine Learning in Materials Science

The DOI matches its Crossref record; 69 of the 73 references it cites were found in a registry or an index, and 4 are not checked yet.

Model agnostic generation of counterfactual explanations for molecules

Open access
Wellawatte, Geemi P., Seshadri, Aditi, White, Andrew D.·2022·Chemical Science
DOI 10.1039/d1sc05259dPMC PMC8966631

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

Generating model agnostic molecular counterfactual explanations to explain model predictions.

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 · 10.1039/d1sc05259d

Cites

69 of 73 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 An example of a counterfactual. The molecule on left was predicted to have class of 0, no activity. With the modification shown in teal, the molecule would be in class 1, active. This shows that the carboxylic acid is an explanation for lack of activity.
Fig. 1 An example of a counterfactual. The molecule on left was predicted to have class of 0, no activity. With the modification shown in teal, the molecule would be in class 1, active. This shows that the carboxylic acid is an explanation for lack of activity.CC BY · PubMed Central
Fig. 2 Overview of MMACE. The input is a molecule to be predicted. Chemical space is expanded and clustered. Counterfactuals are selected from clusters to find succinct explanation of base molecule prediction.
Fig. 2 Overview of MMACE. The input is a molecule to be predicted. Chemical space is expanded and clustered. Counterfactuals are selected from clusters to find succinct explanation of base molecule prediction.CC BY · PubMed Central
Fig. 3 Counterfactual for negative example of blood–brain barrier random forest model. Similarity is computed from Tanimoto similarity of ECFP4 fingerprints.65 Red indicates deletion relative to base molecule and teal indicates modification. Counterfactuals show that the removing or modifying carboxylic acid group is the simplest way to make this molecule pass the blood–brain barrier.
Fig. 3 Counterfactual for negative example of blood–brain barrier random forest model. Similarity is computed from Tanimoto similarity of ECFP4 fingerprints.65 Red indicates deletion relative to base molecule and teal indicates modification. Counterfactuals show that the removing or modifying carboxylic acid group is the simplest way to make this molecule pass the blood–brain barrier.CC BY · PubMed Central
Fig. 4 Chemical space for solubility predicting RNN model. This is a principle component analysis of chemical space from Tanimoto similarity distances. Points are colored by solubility. Counterfactuals are annotated.
Fig. 4 Chemical space for solubility predicting RNN model. This is a principle component analysis of chemical space from Tanimoto similarity distances. Points are colored by solubility. Counterfactuals are annotated.CC BY · PubMed Central
Fig. 5 Counterfactuals for positive example of GCN model for classifying HIV activity. Similarity is computed from Tanimoto similarity of ECFP4 fingerprints.65 Teal indicates the modifications to the base molecule. Counterfactuals illustrate which modifications make the base molecule HIV active.
Fig. 5 Counterfactuals for positive example of GCN model for classifying HIV activity. Similarity is computed from Tanimoto similarity of ECFP4 fingerprints.65 Teal indicates the modifications to the base molecule. Counterfactuals illustrate which modifications make the base molecule HIV active.CC BY · PubMed Central
Fig. 6 The effect of mutation number on Tanimoto similarity of generated molecules from RNN solubility model. Increasing mutation number reduces number of similar molecules from which counterfactuals can be generated.
Fig. 6 The effect of mutation number on Tanimoto similarity of generated molecules from RNN solubility model. Increasing mutation number reduces number of similar molecules from which counterfactuals can be generated.CC BY · PubMed Central
Fig. 7 The effect of STONED alphabet choice counterfactuals from RNN solubility model. Although each counterfactual has the same similarity, the molecules are increasingly unusual. The basic alphabet provides a balance of intuitive counterfactuals and enough tokens to explore chemical space.
Fig. 7 The effect of STONED alphabet choice counterfactuals from RNN solubility model. Although each counterfactual has the same similarity, the molecules are increasingly unusual. The basic alphabet provides a balance of intuitive counterfactuals and enough tokens to explore chemical space.CC BY · PubMed Central
Fig. 8 PubChem53 derived counterfactuals from the blood–brain barrier permeation prediction.
Fig. 8 PubChem53 derived counterfactuals from the blood–brain barrier permeation prediction.CC 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 67 references

Research works printed but not linked: 22

  1. Huang, 2016, Nat. Commun., 7, 1
  2. Lee, 2004, Hum. Factors, 46, 50, 10.1518/hfes.46.1.50.30392 · DOI · Paper page
  3. Goodman, 2017, AI Mag., 38, 50
  4. A. I. ACT , On Artificial Intelligence: A European Approach to Excellence and Trust , 2021 , COM/2021/206
  5. O.Biran and C.Cotton , IJCAI-17 workshop on explainable AI (XAI) , 2017 , pp. 8–13
  6. M.Sundararajan , A.Taly and Q.Yan , International Conference on Machine Learning , 2017 , pp. 3319–3328
  7. D.Erhan , Y.Bengio , A.Courville and P.Vincent , Technical Report , Univeristé de Montréal , 2009
  8. P. W.Koh and P.Liang , International Conference on Machine Learning , 2017 , pp. 1885–1894
  9. Humer, 2021, ChemRxiv
  10. M. F.Frisch , Theories, models, and explanation , University of California , Berkeley , 1998
  11. M. A.Bruckner , Banking & Financial Services Policy Report , 2018 , vol. 37
  12. S.Singla , Machine Learning to Predict Credit Risk in Lending Industry , 2020 , https://www.aitimejournal.com/@saurav.singla/machine-learning-to-predict-credit-risk-in-lending-industry
  13. Wachter, 2017, Harv. J. Law Technol., 31, 841
  14. J.Woodward , The Stanford Encyclopedia of Philosophy, Metaphysics Research Lab , Stanford University , Winter , 2016th edn, 2016
  15. Mauri, 2006, Match, 56, 237
  16. P. W.Battaglia , J. B.Hamrick , V.Bapst , A.Sanchez-Gonzalez , V.Zambaldi , M.Malinowski , A.Tacchetti , D.Raposo , A.Santoro , R.Faulkner et al. , 2018 , arXiv preprint arXiv:1806.01261 · arXiv
  17. T. T.Tanimoto , Internal IBM Technical Report , 1958
  18. M.Ester , H.-P.Kriegel , J.Sander and X.Xu , KDD , 1996 , pp. 226–231
  19. Pedregosa, 2011, J. Mach. Learn. Res., 12, 2825
  20. Savjani, 2012, Int. Scholarly Res. Not., 2012, 195727
  21. Bradley, 2010, Nat. Preced., 1
  22. D.Kingma and J.Ba , International Conference on Learning Representations , 2014

Cited by

Sorted by citations

Related

See all

Metadata

Authors
Wellawatte, G. P., Seshadri, A., & White, A. D.
Affiliations
University of Rochester (US)
Corresponding author
White, Andrew D.
ORCID
Venue
Chemical Science
Publisher
Royal Society of Chemistry (RSC)
Published
2022
Volume
13
Issue
13
Pages
3697-3705
ISSN
  • 2041-6520 (ISSN-L)
  • 2041-6539
DOI
10.1039/d1sc05259d
PMC
PMC8966631
OpenAlex
W4225876201
Cited by
143
References
67
Citations by year
2026: 32 · 2025: 27 · 2024: 28 · 2023: 34 · 2022: 22
FWCI
9.13
Type
journal article
Language
en
Open access
Licence
Locations
  • Chemical Science (publishedVersion, cc-by)
  • ChemRxiv (acceptedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
  • National Institute of General Medical Sciences (R35GM137966)
  • National Science Foundation (1764415)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
  • Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
Keywords
counterfactual explanations, model-agnostic explanation, explainable deep learning, interpretability, black-box models, molecular property prediction, graph neural networks, random forest, sequence models, spurious correlations, molecular structure

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