The DOI matches its Crossref record; 171 of the 171 references it cites were found in a registry or an index; 80 of the 149 sources its claims cite were checked for support.
Machine Learning for the Discovery, Design, and Engineering of Materials.
Open accessAbstract
Machine learning (ML) has become a part of the fabric of high-throughput screening and computational discovery of materials. Despite its increasingly central role, challenges remain in fully realizing the promise of ML. This is especially true for the practical acceleration of the engineering of robust materials and the development of design strategies that surpass trial and error or high-throughput screening alone. Depending on the quantity being predicted and the experimental data available, ML can either outperform physics-based models, be used to accelerate such models, or be integrated with them to improve their performance. We cover recent advances in algorithms and in their application that are starting to make inroads toward ( a) the discovery of new materials through large-scale enumerative screening, ( b) the design of materials through identification of rules and principles that govern materials properties, and ( c) the engineering of practical materials by satisfying multiple objectives. We conclude with opportunities for further advancement to realize ML as a widespread tool for practical computational materials design.
Overview
CoChat Summary
TL;DR
This review surveys how machine learning (ML) is being integrated into computational materials science across three overlapping activities: discovery via accelerated high-throughput screening, design via inferred structure-property relationships, and engineering via multi-objective optimization. It argues ML can outperform, accelerate, or complement physics-based models depending on the property and data available, and it catalogs remaining challenges in data bias, uncertainty, and transferability.
Key findings
- ML regression models can accelerate screening of enumerated materials spaces from hundreds (trial-and-error) to thousands or millions of compounds; enumerated sets like GDB-17 (66 billion organic molecules), OHLDB, hMOF, ToBaCCo, CoRE-MOF (>14,000 MOFs), and tmQM (86,000 structures) enable this.
- Nandy et al. trained an ANN on an 80/20 split of 712 open-shell transition-metal complexes to predict metal-oxo formation energetics to 5.5 kcal/mol test accuracy, then predicted ~10k catalysts, revealing strong metal- and spin-state dependence of terminal metal-oxo stability.
- Liu et al. applied an ANN predicting the multireference diagnostic r_ND to 3% mean absolute percent error (0.018) over 187.2k new complexes (only 1% seen in training), finding that higher multireference character in low-spin states holds only for some metals (e.g., Cr, Mn).
- Incorporating OHLDB compounds reduced ANN errors on experimental (CSD) complexes by approximately 25%; Janet et al.'s EGO multi-objective optimization of 2.8M candidate TMCs improved solubility and redox potential at least 500-fold faster than random/diversity-oriented search, achieving leads in weeks instead of decades.
Methods
- Narrative review (not a primary study) covering ML algorithms (KRR, GP, ANN/CNN, random forest, LASSO, genetic algorithms, active learning/EGO) and their application to transition-metal complexes and metal-organic frameworks, with some coverage of organic molecules and experimental data sets.
- Organizes the field into three framings—discovery (enumerative screening), design (structure-property relationships and feature selection), and engineering (single- and multi-objective optimization)—illustrated with cited demonstrations and figures.
- Discusses feature selection/engineering approaches (revised autocorrelations, SHAP, grad-CAM, recursive feature addition, dimensionality reduction such as PCA, t-SNE, UMAP) and their role in interpreting models.
Limitations
- Stated by the paper: enumeration rules and training data can introduce bias; DFT screening can fail to converge on hypothetical materials, leaving ML models least reliable where calculations fail; experimental data sets suffer positive publication bias toward successful experiments and inconsistent reporting; KRR models trained on different MOF data sets (CoRE-MOF vs. BW-DB/ToBaCCo) were not transferable; GP scaling is supralinear with data set size; generative models can produce experimentally infeasible compounds and inherit data biases; a hierarchy of tractable methods remains out of reach for correlated materials like TMCs; structure may be unknown or evolve during a material's operative life cycle.
- Evident from the text: this is a review, so no new primary data or experiments are reported; many quantitative results are drawn from cited prior work rather than independently validated here.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
40 Valid · 5 Not valid · 19 Not checked · 39 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The largest data set of experimentally characterized metal-organic compounds (e.g., MOFs or TMCs) is the set of structures obtained from X-ray diffraction in the Cambridge Structural Database (CSD) (25).”
§ Machine Learning for the Discovery, Design, and Engineering of Materials > FEATURE SELECTION AND ENGINEERING
What the sources say
- Partly supportedcites [25] sourceExperimental — accuracy not yet measured
In “The value of sharing crystal structures” (page unavailable), the paper says the CSD contains all published organic and metal–organic small-molecule crystal structures, including X-ray diffraction analyses. It does not establish that these structures are the largest dataset of all experimentally characterized metal–organic compounds, and it notes that the CSD also includes neutron diffraction analyses. (LLM, conf=0.85)
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
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“in cases where no physics-based modeling protocol is established, ML can be used to bring computational materials science closer to counterpart experimental efforts, such as by predicting quantities related to synthesis (12)(13)(14)(15)(16) or materials stability (17).”
§ Machine Learning for the Discovery, Design, and Engineering of Materials > 1. INTRODUCTION
What the sources say
- Supportedcites [17] source
Split into 2 atoms — 2 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
- Not matchedcites (12)(13)(14)(15)(16)
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plainstates a possible 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.
- Supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“As a result, the scale of new materials that can be screened for discovery has been transformed from hundreds by trial and error to thousands or millions.”
§ Machine Learning for the Discovery, Design, and Engineering of Materials > 2. DISCOVERY THROUGH ACCELERATED HIGH-THROUGHPUT SCREENING
What the sources say
- Not matchedcites (1)(2)(3)(4)(19)(20)(21)(22)(23)(24)
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Could not verifycites [25] sourceExperimental — accuracy not yet measured
The Cambridge Structural Database section reports more than 800,000 recorded structures, and ‘Using the CSD > Accessibility and efficiency’ describes tools for searching them. Neither passage says that discovery screening expanded from hundreds of trial-and-error candidates to thousands or millions; searches for screening scale and trial and error found no such account. (LLM, conf=0.75)
- Not matchedcites (9,(26)(27)(28)(29)(30)
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
Every source the paper cites for this claim was checked and supports it.
In the paper
“When KRR models are applied with tailored feature sets, their prediction accuracy can rival or surpass (22) deep learning models (e.g., fully connected ANNs or convolutional neural networks).”
§ Machine Learning for the Discovery, Design, and Engineering of Materials > REGULARIZED MULTIPLE LINEAR REGRESSION
What the sources say
- Supportedcites [22] source
The Model Performance section reports that feature-selected KRR had nearly the same test error as an ANN and outperformed it on the OH64 set. The Methods section describes the ANN as having two fully connected hidden layers, supporting the comparison with a deep learning model. (LLM, conf=0.90)
plainstates an association- 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.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Moosavi et al. (40) noted that KRR models trained to predict gas separation characteristics (e.g., CO 2 adsorption) on different data sets [e.g., experimental CoRE-MOF (29) versus hypothetical BW-DB (38) or ToBaCCo (28)] were not transferable.”
§ Machine Learning for the Discovery, Design, and Engineering of Materials > 2.3. Outstanding Challenges for Enumeration in Transition-Metal Chemistry
What the sources say
- Partly supportedcites [40] sourceExperimental — accuracy not yet measured
In Results, “Applications of diversity analysis” (page not provided), the paper reports poor cross-database transferability for models trained on biased datasets, but those models are random forests, not KRR. In “Predicting adsorption properties of MOFs” (page not provided), it reports good KRR predictions for adsorption properties within the CoRE-2019 and BW-20K databases; it does not establish that KRR models failed to transfer between them. (LLM, conf=0.88)
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
42 more claims · 19 with their cited sources checked
Create a free account to see all 103 claims and to verify the 1 reference still open.
- It may state two separate claims as one. 34
- Probably not a substantial claim. 36
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
§ Machine Learning for the Discovery, Design, and Engineering of Materials > Kernel ridge regression (KRR):
“The GDB-17 (35) set of organic molecules of up to 17 heavy atoms with a limited number of elements (i.e., C, N, O, S and halogens) contains 66 billion compounds”
§ Machine Learning for the Discovery, Design, and Engineering of Materials > References
“35. Ruddigkeit L, van Deursen R, Blum LC, Reymond J-L. 2012. Enumeration of 166 billion organic small molecules in the Chemical Universe Database GDB-17.”
Citations
Cites
Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.74 · Cited by 24,513
A Unified Approach to Interpreting Model Predictions
Lundberg, S., & Lee, S.-I. (2017) · arXiv · doi:10.48550/arxiv.1705.07874 · Cited by 23,707
Commentary: The Materials Project: A materials genome approach to accelerating materials innovation
Jain, A., Ong, S. P., Hautier, G., Chen, W., Richards, W. D., Dacek, S., Cholia, S., Gunter, D., Skinner, D., Ceder, G., et al. (2013) · APL Materials · doi:10.1063/1.4812323 · Cited by 13,094
The Cambridge Structural Database
Groom, C. R., Bruno, I. J., Lightfoot, M. P., & Ward, S. C. (2016) · Acta Crystallographica Section B Structural Science, Crystal Engineering and Materials · doi:10.1107/s2052520616003954 · Cited by 10,297
UMAP: Uniform Manifold Approximation and Projection
McInnes, L., Healy, J., Saul, N., & Großberger, L. (2018) · Journal of Open Source Software · doi:10.21105/joss.00861 · Cited by 9,826
UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
McInnes, L., Healy, J., & Melville, J. (2018) · arXiv · doi:10.48550/arxiv.1802.03426 · Cited by 7,101
Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2020) · International Journal of Computer Vision · doi:10.1007/s11263-019-01228-7 · Cited by 6,909
Steric effects of phosphorus ligands in organometallic chemistry and homogeneous catalysis
Tolman, C. A. (1977) · Chemical Reviews · doi:10.1021/cr60307a002 · Cited by 4,824
Proceedings of the 21st International Conference on Neural Information Processing Systems
Koller, D., Schuurmans, D., Bengio, Y., & Bottou, L. (2008) · Cited by 4,461
MoleculeNet: a benchmark for molecular machine learning
Wu, Z., Ramsundar, B., Feinberg, E. N., Gomes, J., Geniesse, C., Pappu, A. S., Leswing, K., & Pande, V. (2018) · Chemical Science · doi:10.1039/c7sc02664a · Cited by 3,107
Research works printed but not linked: 2
- van der Maaten L, 2008, J. Mach. Learn. Res., 9, 2579
- Moreau G, 1980, Nouv. J. Chim., 4, 359
Cited by
AI for Nanomaterials Development in Clean Energy and Carbon Capture, Utilization and Storage (CCUS)
Chen, H., Zheng, Y., Li, J., Li, L., & Wang, X. (2023) · ACS Nano · doi:10.1021/acsnano.3c01062 · Cited by 132
Advances in data‐assisted high‐throughput computations for material design
Xu, D., Zhang, Q., Huo, X., Wang, Y., & Yang, M. (2023) · Materials Genome Engineering Advances · doi:10.1002/mgea.11 · Cited by 116
Machine learning for analyses and automation of structural characterization of polymer materials
Lu, S., & Jayaraman, A. (2024) · Progress in Polymer Science · doi:10.1016/j.progpolymsci.2024.101828 · Cited by 47
Machine learning guided hydrothermal synthesis of thermochromic VO2 nanoparticles
Chen, Y., Ji, H., Lu, M., Liu, B., Zhao, Y., Ou, Y., Wang, Y., Tao, J., Zou, T., Huang, Y. H., et al. (2023) · Ceramics International · doi:10.1016/j.ceramint.2023.07.035 · Cited by 39
Duan, C., Nandy, A., Terrones, G., Kastner, D. W., & Kulik, H. J. (2022) · JACS Au · doi:10.1021/jacsau.2c00547 · Cited by 40
Self-citation
Metadata
- Authors
- Duan, C., Nandy, A., & Kulik, H. J.
- Affiliations
- Massachusetts Institute of Technology (US)
- Venue
- Annual Review of Chemical and Biomolecular Engineering
- Publisher
- Annual Reviews
- Published
- 2022-06-10
- Volume
- 13
- Issue
- 1
- Pages
- 405-429
- ISSN
- 1947-5438 (ISSN-L)
- 1947-5446
- OpenAlex
- W4220658830
- Cited by
- 69
- References
- 153
- Citations by year
- 2026: 24 · 2025: 19 · 2024: 13 · 2023: 11 · 2022: 2
- FWCI
- 0.73
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Annual Review of Chemical and Biomolecular Engineering (publishedVersion)
- PubMed (publishedVersion)
- Funders
- National Science Foundation (1846426, 1122374, 1704266)
- U.S. Department of Energy (DESC0012702, DESC0018096)
- Defense Advanced Research Projects Agency (D18AP00039)
- Office of Naval Research (N00014-17-1-2956, N00014-17-1, N00014-18-1-2434, N00014-20-1-2150, N00014)
- Division of Chemical, Bioengineering, Environmental, and Transport Systems (CBET-1846426, CBET-1704266)
- Topics
- Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
- Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
- Physical Sciences › Physics and Astronomy › Atomic and Molecular Physics, and Optics › Advanced Chemical Physics Studies
- Keywords
- machine learning, materials discovery, materials design, materials engineering, high-throughput screening, computational materials design, physics-based modeling, multi-objective optimization, materials property prediction
- MeSH terms
- Machine Learning, Algorithms, High-Throughput Screening Assays
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: Machine Learning for the Discovery, Design, and Engineering of Materials. https://app.cochat.ai/papers/10.1146/annurev-chembioeng-092320-120230
- The paper
- Duan, C., Nandy, A., & Kulik, H. J. (2022). Machine Learning for the Discovery, Design, and Engineering of Materials. Annual Review of Chemical and Biomolecular Engineering. https://doi.org/10.1146/annurev-chembioeng-092320-120230
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.