The DOI matches its Crossref record; 197 of the 197 references it cites were found in a registry or an index; 55 of the 90 sources its claims cite were checked for support.
Artificial Intelligence for Drug Discovery: Are We There Yet?
Open accessAbstract
Drug discovery is adapting to novel technologies such as data science, informatics, and artificial intelligence (AI) to accelerate effective treatment development while reducing costs and animal experiments. AI is transforming drug discovery, as indicated by increasing interest from investors, industrial and academic scientists, and legislators. Successful drug discovery requires optimizing properties related to pharmacodynamics, pharmacokinetics, and clinical outcomes. This review discusses the use of AI in the three pillars of drug discovery: diseases, targets, and therapeutic modalities, with a focus on small-molecule drugs. AI technologies, such as generative chemistry, machine learning, and multiproperty optimization, have enabled several compounds to enter clinical trials. The scientific community must carefully vet known information to address the reproducibility crisis. The full potential of AI in drug discovery can only be realized with sufficient ground truth and appropriate human intervention at later pipeline stages.
Overview
CoChat Summary
TL;DR
This review surveys AI for drug discovery across diseases, targets, and therapeutic modalities, focusing on small molecules, and finds AI has enabled compounds to enter clinical trials but no approved medicine can yet be attributed to AI. It concludes that AI4DD is not fully automated and that realizing its potential requires ground truth, reproducibility, and human intervention.
Key findings
- AI4DD publications rose from 49 in 2011 to 333 in 2020; AI-designed compounds have entered trials, including DSP-1181 (phase I, Jan 2020), DSP-0039 (phase I, May 2021), ISM01-055/ISM018-055 (phase II, June 2023), and RLY-1971 (phase I).
- KG-based XGBoost models: the AD model used 53 positive and 3,952 negative genes and identified five previously unassociated immunity genes (FRRS1, CTRAM, SCGB3A1, CIBAR2, TMEFF2); the ATG model used 103 positives and 3,468 negatives, with 43 of the top 251 already in the Autophagy Database, 15 with potential ATG associations, and 193 (77%) “ATG dark genes”; 7 of the top 20 and 2 bottom-predicted genes were validated by literature.
- Generative chemistry: GENTRL produced 30,000 compounds, filtered to six novel DDR1/DDR2 inhibitor candidates; two were potent nM inhibitors, and the process from DDR1 nomination to Compound 6 took 35 days; Compound 1 was orally bioavailable in mice with a 3.5-hour half-life after a 15 mg/kg dose.
- Challenges: no regulatory-approved medicine can be attributed to AI; ML models are often trained on tens of thousands of compounds versus 30 billion virtual compounds; Bayer and Amgen reported 33% and 11% reproducibility rates; data leakage was detected in 15 of 20 biology/medicine subject areas.
Methods
- Design: narrative review of AI for drug discovery (AI4DD), focused primarily on small molecule drugs, organized around three pillars—diseases, targets, and therapeutic modalities—plus regulatory science and caveats.
- Data/sample: cited literature, public databases/resources (e.g., DrugCentral, Pharos, Therapeutic Data Commons, OCHEM), and case studies of AI-designed compounds; no new primary experimental data are reported.
- Analysis: qualitative synthesis and discussion of cited examples; no systematic search strategy, meta-analysis, or formal quantitative synthesis is stated.
Limitations
- Paper-stated: no approved medicine can be attributed to AI, and no fully automated AI-driven drug discovery system exists; most successes rely on computational platforms supporting human decision-making.
- Paper-stated: ML models face applicability domain, external predictivity validation, and chemical diversity concerns; KG-based ML lacks ground truth and can suffer data leakage; reproducibility crisis affects both experiments and ML.
- Evident from text: as a narrative review, it does not report a systematic search or quantitative synthesis, and its examples are illustrative rather than exhaustive.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures

![Figure 2 Chemical structure of two epithelial discoidin domain-containing receptor inhibitors (compounds 1 and 2), developed using the GENTRL (118) generative chemistry approach. Their chemical similarity with ponatinib (bottom left) is discussed in the text. The structure and bioactivity of MMV1803022, one of the community-driven anti-malarial AI competition compounds, is shown bottom center. Pfal summarizes the in vitro growth inhibition of asexual blood-stage P. falciparum (3D7) parasite maturation; Pf ATP4 indicates confirmed on-target biochemical (cytosolic [Na+] increase) inhibition. Halicin is on the bottom right (see text for details). Bioactivities were extracted from (118) for Compounds 1 and 2, from DrugCentral (47) for ponatinib, and from (119) for the anti-malarial compound. CIDs are PubChem compound identifiers.](https://figures.cochat.ai/paper-figures/d2724807da564489bcd96336c2a37301/artificial-intelligence-for-drug-discovery-are-we-there-yet-figure-2.png)


Verification
Claims the paper makes
19 Valid · 24 Not valid · 24 Not checked · 20 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“Rigorous analyses suggest that GCNNs do not improve predictive performance compared to 'classical QSAR' fingerprints (92).”
What the sources say
- Disputedcites [92] sourceExperimental — accuracy not yet measured
In “External Dataset Validation: MUV” (page unavailable), the paper reports that Graph-CNN had the best performance among five methods on four of eight targets, while the ligand-based AP method led on the other four. Its analysis says Graph-CNNs outperformed ligand-based methods on the harder cases, rather than finding that they do not improve predictive performance over fingerprint-based methods. (LLM, conf=0.88)
Checked by a reasoning model — may be wrong.
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“The state-ofthe-art in drug indication curation from text was expanded through InContext (49), which includes therapeutic context information about drug usage, especially for antineoplastic and cardiovascular drugs.”
What the sources say
- Could not verifycites [49] sourceExperimental — accuracy not yet measured
The Previous Efforts and ontology-based formalization sections discuss therapeutic context and text-mining tools, but do not identify InContext or describe its contribution. Searches for InContext, antineoplastic drugs, and cardiovascular drugs found no passage substantiating the attributed claim. (LLM, conf=0.74)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“An ML model based on convolutional neural networks (CNN) has demonstrated near-physician accuracy in diagnosing atopic dermatitis.”
What the sources say
- Partly supportedcites [43] sourceExperimental — accuracy not yet measured
The Abstract says deep neural networks have achieved human-level performance in various image-based medical diagnosis tasks and gives a dermatologist-level skin cancer classifier as an example. It does not report near-physician accuracy for diagnosing atopic dermatitis; searches for atopic dermatitis and eczema found no such passage. The paper’s own case study concerns breast cancer screening. (LLM, conf=0.85)
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“Automated systems that digest large sets of data using named entity recognition, NER (18), are an integral part of public domain databases, for example, DISEASES for gene-disease associations (19); STRING for protein-protein interactions (20); and OpenTargets (21) and Pharos (22) for complex disease-protein-drug annotations, to name a few.”
What the sources say
- Not checked — no full text of the sourcecites [18] source
- Not matchedcites [19] Borthwick A, 1998
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites [20] source
Citation 19 is attached specifically to DISEASES as a database of gene–disease associations, not to the other named databases. The Named entity recognition section describes DISEASES using NER to process large text corpora, and the Conclusion identifies it as a freely available resource of disease–gene associations. (LLM, conf=0.94)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Although DrugCentral maps drug uses to existing terminologies like SNOMED-CT and MeSH, two global standards for health terms, capturing therapeutic intent computationally is challenging (48).”
What the sources say
- Partly supportedcites [48] sourceExperimental — accuracy not yet measured
The Web interface section (page unavailable) says DrugCentral supports drug-use searches through SNOMED-CT mappings. The Pharmacological classification section (page unavailable) reports MeSH mappings for drug classes, not specifically for drug uses. The Summary and future directions section says capturing veterinary extra-label uses is “not trivial,” but does not make the broader claim that computationally capturing therapeutic intent is challenging. (LLM, conf=0.86)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
34 more claims · 17 with their cited sources checked
Create a free account to see all 87 claims and to verify the 3 references still open.
- Probably not a substantial claim. 36
- It may state two separate claims as one. 20
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.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3.2. Target identification and validation · p. 7
“In this case, XGBoost, a form of extreme gradient boosting (67), was utilized to identify novel genes associated with Alzheimer's disease (68).”
§ 8. References · p. 25
“67. Köhler S, Carmody L, Vasilevsky N, Jacobsen JOB, Danis D, et al. 2019. Expansion of the Human Phenotype Ontology (HPO) knowledge base and resources. Nucleic Acids Res . 47(D1):D1018-D1027”
Cited works that were retracted
Works this paper cites that were later retracted, from the retraction notices publishers deposit. Major when the notice came out before this paper was published.
A note, not an issue: for example, a cited work that was retracted only after this paper was published, which its authors could not have known.
In the paper
“Diseases and healthcare: The last decade has witnessed significant computational advances in disease diagnosis, with ten or more publications describing the use of AI in dermatology, tuberculosis, Alzheimer’s disease, diabetes, hypertension, and cancer, reviewed elsewhere (32).”
Cited work 10.1007/s12652-021-03612-z · retracted 2026-03-09 · notice 10.1007/s12652-026-05063-w · this paper published 2024-01-23
Citations
Cites
1 retracted
Kumar, Y., Koul, A., Singla, R., & Ijaz, M. F. (2023) · Journal of Ambient Intelligence and Humanized Computing · doi:10.1007/s12652-021-03612-z · Cited by 1,040 · cited 1×
“Diseases and healthcare: The last decade has witnessed significant computational advances in disease diagnosis, with ten or more publications describing the use of AI in dermatology, tuberculosis, Alzheimer’s disease, diabetes, hypertension, and cancer, reviewed elsewhere (32).”
BackgroundRetractedXGBoost: A Scalable Tree Boosting System
Chen, T., & Guestrin, C. (2016) · Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2939672.2939785 · Cited by 54,569 · cited 1×
“Centered on the main node, “Target”, nodes listed clockwise are as follows: Tissue; Cell_type; Panther, Protein Analysis Through Evolutionary Relationships, a protein classification scheme(57); Reactome, a pathways database (58); Jensen DISEASES (19); OMIM, Online Mendelian Inheritance in Man database(59); HPO, human…”
BackgroundHighly accurate protein structure prediction with AlphaFold
Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., et al. (2021) · Nature · doi:10.1038/s41586-021-03819-2 · Cited by 43,764 · cited 2×
“Highly accurate protein structure prediction with AlphaFold.”
Background(2017) · Cited by 11,625
Lipinski, C. A., Lombardo, F., Dominy, B. W., & Feeney, P. J. (1997) · Advanced Drug Delivery Reviews · doi:10.1016/s0169-409x(96)00423-1 · Cited by 11,385
Why Most Published Research Findings Are False
Ioannidis, J. P. A. (2005) · PLoS Medicine · doi:10.1371/journal.pmed.0020124 · Cited by 10,505
Szklarczyk, D., Kirsch, R., Koutrouli, M., Nastou, K., Mehryary, F., Hachilif, R., Gable, A. L., Fang, T., Doncheva, N. T., Pyysalo, S., et al. (2023) · Nucleic Acids Research · doi:10.1093/nar/gkac1000 · Cited by 9,726 · cited 4×
“Automated systems that digest large sets of data using named entity recognition, NER (18), are an integral part of public domain databases, for example, DISEASES for gene-disease associations (19); STRING for protein-protein interactions (20); and OpenTargets (21) and Pharos (22) for complex disease-protein-drug annot…”
BackgroundInfluentialTensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
Abadi, M., Agarwal, A., Barham, P., Brevdo, E., Chen, Z., Citro, C., Corrado, G. S., Davis, A., Dean, J., Devin, M., et al. (2016) · arXiv · doi:10.48550/arxiv.1603.04467 · Cited by 9,408
Scikit-learn: Machine Learning in Python
(2011) · Journal of Machine Learning Research · doi:10.5555/1953048.2078195 · Cited by 8,539
Weininger, D. (1988) · Journal of Chemical Information and Computer Sciences · doi:10.1021/ci00057a005 · Cited by 7,436 · cited 3×
“Page | 11 Traditionally, QSARs and other small-molecule property ML models use chemical structures as input, often encoded as SMILES (Simplified Molecular Input Line Entry System) (88).”
MethodInfluential
Research works printed but not linked: 24
- Harris LA., 2021, Artificial intelligence: background, selected issues, and policy considerations
- Borthwick A, 1998, Proceedings of the Sixth Workshop on Very Large Corpora
- Bartusiak M., 1981, Discover Magazine, 47
- Bylinsky G., 1981, Fortune Magazine, 106
- Jassal B, 2020, Nucleic Acids Res., 48, D498
- Hansch C, 1995, Exploring QSAR: Fundamentals and Applications in Chemistry and Biology
- Natl. Cancer Inst, 2023, NCI-60 growth inhibition data
- Deb K, 2006, Int. J. Comput. Intell. Res., 2, 273
- Landrum G., 2022, RDKit.
- Pedregosa F, 2011, J. Mach. Learn. Res., 12, 2825
- Int. Counc. Harmon. Tech. Requir. Pharm. Hum. Use (ICH), 2023, M7: assessment and control of DNA reactive (mutagenic) impurities in pharmaceuticals to limit potential carcinogenic risk
- EMA (Eur. Med. Agency), 2018, Reflection paper on the qualification of non-genotoxic impurities
- Hines PA, 2022, Nat. Rev. Drug Discov.
- FDA (US Food Drug Admin.), 2023, Using artificial intelligence and machine learning in the development of drug and biological products
- FDA (US Food Drug Admin.), 2023, Artifical intelligence in drug manufacturing
- Exscientia, 2020, DSP-1181
- Yoshinaga H, 2020, 2,6-Disubstituted pyridine derivative
- Wills T., 2022, CAS
- Yoshinaga H, 2020, Condensed lactam derivative
- Fultinavičiūtė U., 2023, Clinical Trials Arena
- Aliper A, 2020, Methods of inhibiting kinases
- InSilico Med, 2023, Hong Kong Stock Exchange application proof of InSilico Medicine Cayman TopCo
- Relay Ther, 2020, Relay Therapeutics® announces a worldwide license and collaboration agreement with Genentech for RLY-1971
- OpenAI, 2023, GPT-4 technical report
Cited by
Fu, L., Shi, S., Yi, J., Wang, N., He, Y., Wu, Z., Peng, J., Deng, Y., Wang, W., Wu, C., et al. (2024) · Nucleic Acids Research · doi:10.1093/nar/gkae236 · Cited by 1,260
The Role of AI in Hospitals and Clinics: Transforming Healthcare in the 21st Century
Maleki Varnosfaderani, S., & Forouzanfar, M. (2024) · Bioengineering · doi:10.3390/bioengineering11040337 · Cited by 931
“Exscientia introduced the first AI-designed drug molecule for clinical trials in early 2020 [46].”
BackgroundReinvent 4: Modern AI–driven generative molecule design
Loeffler, H. H., He, J., Tibo, A., Janet, J. P., Voronov, A., Mervin, L. H., & Engkvist, O. (2024) · Journal of Cheminformatics · doi:10.1186/s13321-024-00812-5 · Cited by 319
“Here, we will focus on small molecule design but other modalities are being investigated as well [12, 13].”
Open Targets Platform: facilitating therapeutic hypotheses building in drug discovery
Buniello, A., Suveges, D., Cruz-Castillo, C., Llinares, M. B., Cornu, H., Lopez, I., Tsukanov, K., Roldán-Romero, J. M., Mehta, C., Fumis, L., et al. (2025) · Nucleic Acids Research · doi:10.1093/nar/gkae1128 · Cited by 217
Tribulations and future opportunities for artificial intelligence in precision medicine
Carini, C., & Seyhan, A. A. (2024) · Journal of Translational Medicine · doi:10.1186/s12967-024-05067-0 · Cited by 173
“As discussed in the literature, [27, 28, 39–42], AI-based drug discovery and screening alongside laboratory automation could augment human drug design, chemical synthesis, drug screening, biological testing, and decision-making involved in drug discovery and development, potentially overcoming low…”
BackgroundInfluential
Metadata
- Authors
- Hasselgren, C., & Oprea, T. I.
- Affiliations
- Genentech (US)
- University of New Mexico (US)
- ORCID
- Oprea, Tudor I.
- Venue
- Annual Review of Pharmacology and Toxicology
- Publisher
- Annual Reviews
- Published
- 2024-01-23
- Volume
- 64
- Issue
- 1
- Pages
- 527-550
- ISSN
- 0362-1642 (ISSN-L)
- 1545-4304
- OpenAlex
- W4384389969
- Cited by
- 236
- References
- 167
- Citations by year
- 2026: 67 · 2025: 107 · 2024: 57
- FWCI
- 3.99
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- public-domain
- creativecommons.org/licenses/by/4.0/ (unspecified)
- Locations
- The Annual Review of Pharmacology and Toxicology (publishedVersion, public-domain)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by-nc-sa)
- arXiv (Cornell University)
- Funders
- National Institutes of Health (U24 224370)
- Genentech
- National Cancer Institute (U24 224370)
- 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 › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Keywords
- artificial intelligence in drug discovery, generative chemistry, machine learning, small molecule drugs, therapeutic target, therapeutic modalities, replication crisis, clinical trial candidates, pharmacokinetics, pharmacodynamics
- MeSH terms
- Animals, Artificial Intelligence, Humans, Physicians, Technology, Reproducibility of Results, Drug Discovery
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Artificial Intelligence for Drug Discovery: Are We There Yet? https://app.cochat.ai/papers/10.1146/annurev-pharmtox-040323-040828
- The paper
- Hasselgren, C., & Oprea, T. I. (2024). Artificial Intelligence for Drug Discovery: Are We There Yet? Annual Review of Pharmacology and Toxicology. https://doi.org/10.1146/annurev-pharmtox-040323-040828
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.