CoChat
Sign Up
Cited by 91Machine Learning in Materials ScienceCatalytic Processes in Materials ScienceAdvanced Chemical Physics Studies

Machine Learning Accelerates the Discovery of Design Rules and Exceptions in Stable Metal–Oxo Intermediate Formation

Open access
Nandy, Aditya, Zhu, Jiazhou, Janet, Jon Paul, Duan, Chenru, Getman, Rachel B., Kulik, Heather J.·2019·ACS Catalysis
DOI 10.1021/acscatal.9b02165

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

Metal–oxo moieties are important catalytic intermediates in the selective partial oxidation of hydrocarbons and in water splitting. Stable metal–oxo species have reactive properties that vary depending on the spin state of the metal, complicating the development of structure–property relationships. To overcome these challenges, we train machine-learning (ML) models capable of predicting metal–oxo formation energies across a range of first-row metals, oxidation states, and spin states. Using connectivity-only features tailored for inorganic chemistry as inputs to kernel ridge regression or artificial neural network (ANN) ML models, we achieve good mean absolute errors (4–5 kcal/mol) on set-aside test data across a range of ligand orientations. Analysis of feature importance for oxo formation energy prediction reveals the dominance of nonlocal, electronic ligand properties in contrast to other transition metal complex properties (e.g., spin-state or ionization potential). We enumerate the theoretical catalyst space with an ANN, revealing expected trends in oxo formation energetics, such as destabilization of the metal–oxo species with increasing d-filling, as well as exceptions, such as weak correlations with indicators of oxidative stability of the metal in the resting state or unexpected spin-state dependence in reactivity. We carry out uncertainty-aware evolutionary optimization using the ANN to explore a >37 000 candidate catalyst space. New metal and oxidation state combinations are uncovered and validated with density functional theory (DFT), including counterintuitive oxo formation energies for oxidatively stable complexes. This approach doubles the density of confirmed DFT leads in originally sparsely populated regions of property space, highlighting the potential of ML-model-driven discovery to uncover catalyst design rules and exceptions.

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 · 10.1021/acscatal.9b02165

Cites

131 of 131 references verified0 citation errors · 0 not checkable

Statistics

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?

  • One-shot models such as LASSO or random forests are widely used both as ML models for property prediction and for selection of important features.

    “In a similar fashion, one-shot models such as least absolute shrinkage and selection operator (LASSO) or random forests are widely used both as ML models for property prediction (61) and for selection of important features (22,32,40,43) (Figure 4).”

    From Machine Learning for the Discovery, Design, and Engineering of Materials. (2022) · cites this paper as [43]

    Section 3.2 reports using random forest-ranked recursive feature addition to select features, and Section 4.1 says this identified important descriptors for property prediction. This supports the feature-selection part of the sentence, to which this citation group is attached. The paper describes a recursive procedure, not a one-shot model, and does not discuss LASSO. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • Nandy et al. trained an ANN on an 80/20 train/test split of 712 open-shell TMCs to predict the reaction energetics for metal-oxo formation to a test set accuracy of 5.5 kcal/mol.

    “Nandy et al. (43) trained an ANN on an 80/20 train/test split of 712 openshell TMCs to predict the reaction energetics for metal-oxo formation to a test set accuracy of 5.5 kcal/mol.”

    From Machine Learning for the Discovery, Design, and Engineering of Materials. (2022) · cites this paper as [43]

    Section 3.2 says the paper trained ANN models using an 80/20 train/test partition, and section 4.4 identifies 712 complexes, with 570 in training and 142 in testing. But section 4.1 assigns the 5.5 kcal/mol test mean absolute error to a feature-selected KRR model, not an ANN, so the claimed ANN accuracy is unsupported. (LLM, conf=0.90)

    Experimental — accuracy not yet measured

  • In some cases, deep learning models have been developed to predict frontier orbital energies, which can be applied to explain experimental activity of catalysts known to be correlated to such energies.

    “In some cases, deep learning models (e.g., ANNs) have been developed to predict frontier orbital energies (22,43,86), which can be applied to explain experimental activity of catalysts known to be correlated to such energies (86).”

    From Machine Learning for the Discovery, Design, and Engineering of Materials. (2022) · cites this paper as [43]

    Section 4.2 reports a machine-learning model for predicting β-HOMO energy, and section 4.5 describes using ANNs to screen catalysts by β-HOMO level. The paper does not show that these predictions explain experimental catalyst activity; its Conclusions instead report only a weak correlation between β-HOMO level and oxo-formation energy in its computational data sets. Page numbers were not available in our copy. (LLM, conf=0.86)

    Experimental — accuracy not yet measured

  • Nandy et al. enumerated and predicted properties of a set of nearly 10k catalysts composed of the original 29 ligands in the data set.

    “they enumerated and predicted properties of a set of nearly 10k catalysts composed of the original 29 ligands in the data set (Figure 3).”

    From Machine Learning for the Discovery, Design, and Engineering of Materials. (2022) · cites this paper as [43]

  • Key observations included the strong dependence on metal and spin state of the stability of formed terminal metal-oxos.

    “Key observations included the strong dependence on metal and spin state of the stability of formed terminal metal-oxos (Figure 3).”

    From Machine Learning for the Discovery, Design, and Engineering of Materials. (2022) · cites this paper as [43]

Self-check

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

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

Cites

Flagged first, then by citations

Research works printed but not linked: 6

  1. Muehlhofer, M.; Strassner, T.; Herrmann, W. A. New Catalyst Systems for the Catalytic Conversion of Methane into Methanol. Angew. Chem., Int. Ed. 2002, 41, 1745-1747.
  2. Lim, M. H.; Rohde, J. U.; Stubna, A.; Bukowski, M. R.; Costas, M.; Ho, R. Y.; Munck, E.; Nam, W.; Que, L., Jr. An FeIVO Complex of a Tetradentate Tripodal Nonheme Ligand. Proc. Natl. Acad. Sci. U. S. A. 2003, 100, 3665-70.
  3. Peterson, A. A.; Nørskov, J. K. Activity Descriptors for CO2 Electroreduction to Methane on Transition-Metal Catalysts. J. Phys. Chem. Lett. 2012, 3, 251-258.
  4. Hammer, B.; Nørskov, J. K. Theoretical Surface Science and CatalysisCalculations and Concepts. Adv. Catal. 2000, 45, 71-129.
  5. Kingma, D. P.; Ba, J. L. Adam: A Method for Stochastic Optimization. International Conference for Learning Representations, San Diego, CA, 2015.
  6. Winkler, J. R.; Gray, H. B. Electronic Structures of Oxo-Metal Ions. In Molecular Electronic Structures of Transition Metal Complexes I. Structure and Bonding, Mingos, D., Day, P., Dahl, J., Eds.; Springer, Berlin: Heidelberg, 2011; Vol. 142, pp 17-28.

Cited by

Sorted by citations

Related

Authors
Nandy, A., Zhu, J., Janet, J. P., Duan, C., Getman, R. B., & Kulik, H. J.
Affiliations
  • Massachusetts Institute of Technology (US)
  • Clemson University (US)
Corresponding author
  • Getman, Rachel B.
  • Kulik, Heather J.
ORCID
Venue
ACS Catalysis
Publisher
American Chemical Society (ACS)
Published
2019-07-23
Volume
9
Issue
9
Pages
8243-8255
ISSN
2155-5435 (ISSN-L)
DOI
10.1021/acscatal.9b02165
OpenAlex
W2963312448
Cited by
91
References
131
Citations by year
2026: 4 · 2025: 7 · 2024: 9 · 2023: 10 · 2022: 20 · 2021: 20 · 2020: 17 · 2019: 4
FWCI
5.51
Type
journal article
Language
en
Open access
Licence
Locations
  • ACS Catalysis (publishedVersion, cc-by-nc-nd)
  • ChemRxiv (acceptedVersion, cc-by-nc-nd)
  • OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc-nd)
  • Figshare (submittedVersion, cc-by-nc-nd)
Funders
Basic Energy Sciences (DE-SC0012702)
Topics
  • Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
  • Physical Sciences › Materials Science › Materials Chemistry › Catalytic Processes in Materials Science
  • Physical Sciences › Physics and Astronomy › Atomic and Molecular Physics, and Optics › Advanced Chemical Physics Studies
Keywords
metal-oxo intermediates, 3d transition metals, spin states, structure-property relationships, machine learning prediction, artificial neural networks, kernel ridge regression, feature importance, ligand electronic properties, evolutionary optimization, catalyst discovery, oxidation state, oxidative stability

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