Pith. sign in

REVIEW 3 major objections 5 minor 62 references

Combining feature-based approaches with graph neural networks and symbolic regression for synergistic performance and interpretability

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Hidden GNN features lift transparent MODNet past black-box rivals

desk verdict Useful hybrid featurizer with a real evaluation leak: the headline SOTA-competitive numbers rest on pretrained GNN features that have already seen the MatBench test sets. read the letter →

arxiv 2509.03547 v2 pith:LI6PXSYU submitted 2025-09-02 cond-mat.mtrl-sci cs.LG

Rogério Almeida Gouvêa , Pierre-Paul De Breuck , Tatiane Pretto , Gian-Marco Rignanese , Marcos José Leite Santos This is my paper · ORCID
classification cond-mat.mtrl-scics.LG
keywords MatterVialMODNetgraphneuralnetworksfeature-basedmachinelearningsymbolicregressionSISSOmaterialsinformaticsinterpretable
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that a feature-based materials model, MODNet, can be made competitive with black-box graph neural networks if its ordinary chemical descriptors are supplemented with three extra feature families: latent representations extracted from pretrained GNNs, fast GNN-generated copies of expensive descriptors, and formulas found by symbolic regression. Across the 13-task MatBench benchmark, the additions reduce prediction errors on nearly every task, with reported gains exceeding 40% in several cases, and the augmented model beats the previous record on tasks such as elastic moduli, band gap, metallicity, and formation energy. The point of the hybrid is to keep the transparency of feature-based models while borrowing representational power from deep learning. An integrated interpreter decodes the most important GNN-derived features into short physical formulas, so the accuracy gain is meant to come without losing explainability.

What carries the argument

The load-bearing object is MatterVial, a modular featurizer that turns pretrained graph models into feature generators instead of end-to-end predictors. It supplies three feature families: descriptor-oriented latent features, which are autoencoded MatMiner and orbital-field-matrix descriptors reproduced by a MEGNet proxy; task-oriented latent features, which are activations from MEGNet, ROOST, and ORB layers; and SISSO rung-one formulas built from MatMiner features. The argument then runs through MODNet's feature-selection and deep-ensemble pipeline, which chooses the useful subset of this enlarged pool and thereby reveals which feature groups carry each task. A separate interpreter uses XGBoost surrogates plus SHAP to identify the interpretable descriptors most correlated with each latent feature, and SISSO converts those descriptors into explicit formulas.

What would settle it

Rerun all 13 MatBench tasks with a strictly fold-local protocol that regenerates SISSO formulas inside each training fold and retrains or constrains the pretrained feature extractors on a corpus explicitly excluding every benchmark structure; if the error reductions above 40% shrink toward the MatMiner baseline, the central comparison is optimistic.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that the gap between interpretable feature-based models and end-to-end graph neural networks is not inherent. Feeding MODNet a feature vector that concatenates hidden-layer activations of pretrained structure-based, composition-based, and equivariant models, together with latent encodings of MatMiner and orbital-field-matrix descriptors that a neural network predicts directly from the crystal structure, and paired-feature symbolic-regression formulas, yields consistently lower errors than MatMiner features alone. In the paper's reported numbers, the hybrid reaches or surpasses the best GNN-only results on several MatBench tasks, including the large band-gap and formation-energy datasets, and it sets new feature-based records on tasks where MODNet was already strong. The authors further claim that the most important latent features can be approximated by compact SISSO formulas based on interpretable chemical descriptors, and that feeding those formulas back as new features improves accuracy even further.

Load-bearing premise

The reported benchmark gains assume that every added feature group was generated fairly: per-task SISSO formulas were not selected using held-out labels, and pretrained models' exposure to similar structures did not inflate test scores.

Editorial extensions

If this is right

  • Feature-based models can be treated as the backbone of a materials prediction pipeline; expensive end-to-end architecture search may be unnecessary when pretrained GNN features are available.
  • Computationally heavy descriptors like orbital-field matrices can be replaced by GNN-predicted latent approximations with little or no loss, making large-scale featurization orders of magnitude faster.
  • The interpreter's decoded formulas give researchers a route from model predictions to testable chemical hypotheses, such as which orbital interactions or packing motifs stabilize a perovskite.
  • Feeding interpretable formulas back into the model as hiSISSO features improves accuracy, suggesting that explainability and performance can reinforce each other rather than trade off.
  • On small experimental datasets, where GNNs usually struggle, the hybrid preserves and extends the advantages of feature-based models.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • One natural extension is to regenerate the SISSO formulas inside each cross-validation fold for every task; the paper does not report doing this, and a fold-local version would make the benchmark numbers robust against target leakage.
  • Because the ORB features dominate the gains, substituting other large-scale equivariant machine-learning potentials, or chaining several such models, would directly test whether larger pretraining corpora translate into further error reductions.
  • The same GNN-as-featurizer plus symbolic-decoder template could be carried over to molecular or polymeric datasets, where the interpretable formulas would take the form of fragment or group-contribution rules rather than crystal packing descriptors.
  • If the decoded formulas remain stable across datasets, they could be used proactively as screening rules rather than only as post-hoc explanations.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper introduces MatterVial, an open-source Python framework that augments the feature-based model MODNet with latent representations from pretrained GNNs (MVL/MEGNet, ROOST, ORB), GNN-approximated descriptors (ℓ-MM and ℓ-OFM), and SISSO-derived symbolic-regression features. The authors evaluate the approach on all 13 MatBench v0.1 tasks, reporting error reductions relative to a MatMiner-featurized MODNet baseline and, in several cases, claiming performance competitive with or superior to state-of-the-art end-to-end GNNs. They also present an interpretability module that uses SHAP and SISSO to decode GNN latent features into chemical formulas, and an adjacent-GNN featurizer that trains a task-specific GNN per fold for feature extraction. The central claim is that transparent feature-based models can be made competitive with black-box GNNs while preserving interpretability.

Significance. If the results are leakage-free, this is a practically valuable contribution: the framework is modular and open-source, and the controlled perovskite ablation (Table 2) shows that combining pretrained GNN features with traditional descriptors can substantially improve MODNet. The interpretability module, which translates latent GNN features into explicit formulas, is a useful step toward explainable materials ML. However, the headline claims depend heavily on the three largest Materials-Project-based tasks (matbench_mp_gap, matbench_mp_is_metal, matbench_mp_e_form), where the frozen MVL features are generated by models pretrained on the same MP snapshots that constitute the MatBench test sets. The paper acknowledges this concern in one sentence but does not quantify or exclude the affected results, and the SISSO and feature-preselection steps are not described as per-fold operations. The significance is therefore conditional on a leakage-free evaluation protocol.

major comments (3)
  1. [Results and discussion, Table 1; Methods (MVL MatterVial featurizers); Results caveat] The central claim of superiority over state-of-the-art GNNs is supported in Table 1 primarily by the matbench_mp_gap (37.6% error reduction), matbench_mp_is_metal (77.1%), and matbench_mp_e_form (67.2%) rows. The MVL MEGNet feature extractors were pretrained on the Materials Project snapshots 2018.6.1 (band gap) and 2019.4.1 (formation energy and other targets), which are exactly the corpora from which the MatBench test sets for these three tasks are drawn. For any test object, the frozen MVL features are therefore produced by a model whose weights were fitted with that object's target value; ROOST and ORB pretraining corpora (MP, OQMD, Alexandria, OMat) overlap similarly. The paper's caveat that 'some of the larger MatBench tasks can no longer be considered truly independent test sets' is not quantified, and the abstract and conclusion still report these results as unqualified successes. This is load-bearing: without these rows, the 'superior in several cases' claim loses its main support. The authors should either separate pretraining-overlap tasks from clean tasks, provide a quantitative overlap analysis, or reframe the contribution around the non-overlapping benchmark results.
  2. [Methods (MODNet model training; SISSO-based formula featurizer); Supplementary Table S12] The SISSO formulas are derived once per task from the full MatBench datasets (Supplementary Table S12 lists the same tasks, e.g., matbench_mp_gap, matbench_perovskites, matbench_log_gvrh), and the initial feature preselection for all experiments is described as an XGBoost-based recursive feature elimination to 800 features without stating that it is performed inside each cross-validation fold. If either the SISSO formula selection or the RFE preselection uses test-fold target labels, every result involving SISSO features (Table 1 Best-groups columns; Table 2 step MODNet@ℓ-MM+ℓ-OFM+MVL+SISSO) and potentially all reported feature-selection outcomes are optimistically biased. The sentence 'By opting for the expressions produced at the SIS step instead of the final SISSO formula, versatility and generalization are assured' does not rule out label leakage. Please specify the exact cross-validation nesting (e.g., performing SISSO and RFE on training folds only) or re-evaluate the affected models with a leakage-free protocol.
  3. [Table 2; Results (Synergy of MatterVial features)] The strongest controlled evidence is Table 2's perovskite ablation, which shows a clear improvement from the ORB features (0.0639 to 0.0386 eV/unit cell). However, the incremental gains attributed to SISSO (0.0673 to 0.0653) and to ROOST after SISSO (0.0653 to 0.0639) are within the reported standard deviations of the models, and the final full-feature result (0.0386 ± 0.0009) is only marginally better than the published HackNIP MODNet@ORB baseline (0.0397). The paper's claim that all MatterVial feature groups act synergistically is therefore not convincingly established. Please provide confidence intervals or significance tests for the incremental additions, and state clearly which improvements are meaningful beyond the ORB features.
minor comments (5)
  1. [Table 1 footnote] The footnote formulas for % error reduction are corrupted by Unicode rendering and are not legible; please rewrite with standard mathematical notation.
  2. [Table 1, row Logarithmic Kvrh] The benchmark model is listed as 'coNGN', but the intended model is coGN; fix the typo.
  3. [Methods, MatterVial implementation] The statement that MatterVial reduces feature generation time by 'a minimum of two orders of magnitude' is presented without a benchmark table or measurement description; please add supporting data or soften the claim.
  4. [Abstract; Results caveat] The abstract's 'accuracy increases exceeding 40%' should be qualified by the pretraining-overlap caveat; as written, it conflicts with the Results sentence acknowledging that the larger MatBench tasks are not truly independent test sets.
  5. [Table 1, glass formability footnote] The footnote states that the reported MODNet baseline of 0.960 AUROC could not be replicated; please clarify the exact MODNet configuration and hyperparameters used, and whether the published MODNet results are reproduced for the other tasks.

Circularity Check

2 steps flagged · score 6.0 of 10

Headline gains on MP-based MatBench tasks partly reduce to pretrained-target leakage; SISSO features are target-fitted inputs.

  1. fitted input called prediction [Results and discussion, 'MatBench validation of MatterVial features' (Table 1); Methods, 'MVL MatterVial featurizers']
    "Note, however, that some of the larger MatBench tasks can no longer be considered truly independent test sets for models exposed to vast amounts of similar ab initio data during pretraining. ... we incorporate five pretrained MEGNet models provided by the Materials Virtual Lab. Specifically, these are the models trained for the formation energy, Fermi energy, and elastic constants KVRH and GVRH on the 2019.4.1 Materials Project crystals dataset, as well as the band gap regression model trained on the 2018.6.1 Materials Project crystals dataset."

    The MatBench tasks matbench_mp_gap, matbench_mp_is_metal, and matbench_mp_e_form are constructed from the same Materials Project snapshots used to pretrain the MVL MEGNet feature extractors (106,113 and 132,752 entries in Table 1; SI S4 states the MP2018-stable set is 'Identical to the dataset used for band gap prediction task in MatBench'). For a test-fold crystal, the frozen MVL features are outputs of a model whose weights were fitted to that crystal's target value, so the 'error reductions' of 37.6% (mp_gap), 77.1% (mp_is_metal), and 67.2% (mp_e_form) are partly obtained by feeding the labels through the features. The paper's own caveat admits the test sets are no longer independent, yet those rows still drive the headline claim of superiority over state-of-the-art GNNs.

  2. fitted input called prediction [Methods, 'SISSO-based formula featurizer'; Supplementary Information S8]
    "The SISSO++ framework was used to generate symbolic expressions that approximate target material properties across 15 datasets ... by transforming MatMiner features. ... For our configuration, restricted to rung one, this yields 20 paired-feature formulas. ... These formulas, derived for each of the 15 tasks, are compiled in the file SISSO_FORMULAS_v1.txt, which is accessed by the get_sisso_features function in MatterVial."

    The SISSO formulas are selected by correlation with the target property on the same MatBench datasets (Table S12 lists matbench_perovskites, matbench_mp_gap, matbench_mp_e_form, etc.), and are then appended as input features to MODNet predicting that same target. No per-fold or nested selection procedure is described; the SI configuration leaves out only 5% of the data globally, not the MatBench test fold. The 'symbolically derived' features therefore encode information about the target labels, and the modest improvements attributed to SISSO (e.g., perovskite MAE 0.0653 vs 0.0673) partly reflect a refit of the target through engineered features rather than an independent prediction.

full rationale

The paper is not circular in the sense of a mathematical derivation being equivalent to its assumptions, and it contains substantial independent content: the descriptor-oriented autoencoders (ℓ-MM, ℓ-OFM), the per-fold adjacent GNN featurizer, and the interpretability surrogate pipeline are all self-contained engineering contributions that do not reduce to the benchmark labels. However, two load-bearing evaluation steps do reduce, at least in part, by construction. First, the largest claimed error reductions come from task-oriented pretrained GNN features (MVL, ROOST, ORB) whose pretraining data overlaps the MatBench test sets for mp_gap, mp_is_metal, and mp_e_form; the paper's own caveat admits these tasks are no longer independent, yet the headline numbers still count them. Second, the SISSO-based features are explicitly generated by fitting symbolic expressions to the target properties of those same tasks, with no per-fold selection described, so their contribution is a fitted input rather than an independent predictor. Neither step relies on self-citation chains or imported uniqueness theorems; the cited MODNet and SISSO work is external and methodologically legitimate. The central claim thus has real independent content, but several of the 'prediction' results are statistically forced by target leakage through the features, warranting a partial-circularity score of 6 rather than a clean bill.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

No new physical entities, fields, or forces are introduced; MatterVial is a software framework. The main free parameters are compression ratios, SISSO settings, and feature-selection cutoffs chosen by benchmark performance. The key axiom, and the most fragile one, is that SISSO formula selection does not use test-fold target information.

free parameters (6)
  • Latent OFM compression ratio = 20% (188 features)
    Chosen by comparing predictive MAE in SI Table S1; not derived from first principles.
  • Latent MatMiner compression ratio = 60% (758 features)
    Chosen by comparing predictive MAE in SI Table S2; balances compression and accuracy.
  • Autoencoder hidden layer sizes = 2.5n (OFM), 2.2n (MatMiner)
    Determined by screening reconstruction loss in SI S7.1; architecture is a hand-tuned modeling choice.
  • SISSO configuration = rung=1, desc_dim=2, n_sis_select=10, 20 formulas per task
    Configuration chosen for interpretability and generality; not independently derived.
  • Feature preselection count = 800 features via XGBoost RFE
    Fixed threshold in Methods; affects downstream MODNet feature selection.
  • Surrogate interpreter top-feature count = 30 SHAP-selected descriptors per latent feature
    Hand-chosen cutoff for SISSO input in the interpreter module.
assumptions (4)
  • domain assumption Pretrained GNN features transfer across tasks and databases.
    Central to task-oriented featurizers; tested empirically but not proven, and the paper acknowledges overlap with MatBench data.
  • domain assumption Autoencoder and GNN proxy reconstruction errors are small enough that compressed descriptors preserve useful chemistry.
    Supported by SI Tables S7-S8 and S11, but these are measured on reconstruction, not downstream validity for unseen materials.
  • domain assumption MP2018-stable is an appropriate training pool for descriptor autoencoders and surrogate interpreters.
    Feature generation relies on this Materials Project snapshot and its stability filter.
  • ad hoc to paper SISSO formulas are selected without using test-fold target information.
    The Methods describe SISSO formulas approximating target properties across the same 15 MatBench datasets and do not specify a per-fold or nested selection; if this assumption fails, the SISSO gains are leakage.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Combining feature-based approaches with graph neural networks and symbolic regression for synergistic performance and interpretability." pith.science (2026). https://pith.science/paper/LI6PXSYU

@misc{pith2026250903547,
  author       = {Pith},
  title        = {Pith review of: Combining feature-based approaches with graph neural networks and symbolic regression for synergistic performance and interpretability},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LI6PXSYU}},
  note         = {Machine review of arXiv:2509.03547}
}
read the original abstract

This study introduces MatterVial, an innovative hybrid framework for feature-based machine learning in materials science. MatterVial expands the feature space by integrating latent representations from a diverse suite of pretrained graph neural network (GNN) models including: structure-based (MEGNet), composition-based (ROOST), and equivariant (ORB) graph networks, with computationally efficient, GNN-approximated descriptors and novel features from symbolic regression. Our approach combines the chemical transparency of traditional feature-based models with the predictive power of deep learning architectures. When augmenting the feature-based model MODNet on Matbench tasks, this method yields significant error reductions and elevates its performance to be competitive with, and in several cases superior to, state-of-the-art end-to-end GNNs, with accuracy increases exceeding 40% for multiple tasks. An integrated interpretability module, employing surrogate models and symbolic regression, decodes the latent GNN-derived descriptors into explicit, physically meaningful formulas. This unified framework advances materials informatics by providing a high-performance, transparent tool that aligns with the principles of explainable AI, paving the way for more targeted and autonomous materials discovery.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 50 canonical work pages

  1. [1]

    F., Florea, L., De Oliveira, M

    Rodrigues, J. F., Florea, L., De Oliveira, M. C. F., Diamond, D. & Oliveira, O . N. Big data and machine learning for materials science. Discov Mater 1, 12 (2021)

    work page 2021

  2. [2]

    Dey, A. et al. State of the Art and Prospects for Halide Perovskite Nanocrystals. ACS Nano 15, 10775–10981 (2021)

    2021

  3. [3]

    & Buehler, M

    Guo, K., Yang, Z., Yu, C.-H. & Buehler, M. J. Artificial intelligence and machine learning in design of mechanical materials. Mater. Horiz. 8, 1153–1172 (2021)

    work page 2021

  4. [4]

    & Rignanese, G

    De Breuck, P.-P., Hautier, G. & Rignanese, G. M. Materials property predict ion for limited datasets enabled by feature selection and joint learning with MODNet. npj Computational Materials 7, 1–8 (2021)

    work page 2021

  5. [5]

    & Gao, F

    Zhang, B., Zhou, M., Wu, J. & Gao, F. Predicting the Materials Propertie s Using a 3D Graph Neural Network With Invariant Representation. IEEE Access 10, 62440–62449 (2022)

    work page 2022

  6. [6]

    Tawfik, S. A. & Russo, S. P. Naturally-meaningful and efficient descript ors: machine learning of material properties based on robust one-shot ab initio descriptors. J Cheminform 14, 78 (2022)

    work page 2022

  7. [7]

    Ward, L. et al. Matminer: An open source toolkit for materials data mining. Computational Materials Science 152, 60–69 (2018)

    work page 2018

  8. [8]

    A., Brolo, A

    Pretto, T., Baum, F., Gouvêa, R. A., Brolo, A. G. & Santos, M. J. L. Optimi zing the Synthesis Parameters of Double Perovskites with Machine Learning Us ing a Multioutput Regression Model. J. Phys. Chem. C 128, 7041–7052 (2024)

    work page 2024

Show all 62 references
  1. [9]

    Liu, J. et al. Toward Excellence of Electrocatalyst Design by Emerging Descriptor‐ Oriented Machine Learning. Adv Funct Materials 32, 2110748 (2022)

    2022

  2. [10]

    & Ko, D.-H

    Kim, G.-H., Lee, C., Kim, K. & Ko, D.-H. Novel structural feature-descriptor platform for machine learning to accelerate the development of organic photovoltaics. Nano Energy 106, 108108 (2023)

    2023

  3. [11]

    Li, S. et al. Encoding the atomic structure for machine learning in materials science. WIREs Comput Mol Sci 12, e1558 (2022)

    2022

  4. [12]

    MatBench Leaderboard

    Dunn, A. MatBench Leaderboard. https://matbench.materialsproject.org/ (2024)

    2024

  5. [13]

    Riebesell, J. et al. A framework to evaluate machine learning crystal stability predic tions. Nat Mach Intell 7, 836–847 (2025)

    2025

  6. [15]

    P., Kondor, R

    Bartók, A. P., Kondor, R. & Csányi, G. On representing chemical environments. Phys. Rev. B 87, 184115 (2013)

    2013

  7. [16]

    Kanter, J. M. & Veeramachaneni, K. Deep feature synthesis : Towards automating data science endeavors. in 2015 IEEE International Conference on Data Science and Advanced Analytics (DSAA) 1–10 (IEEE, Campus des Cordeliers, Paris, France, 2015). doi:10.1109/DSAA.2015.7344858

    2015

  8. [17]

    Arik, S. O. & Pfister, T. TabNet: Attentive Interpretable Tabular Le arning. Preprint at https://doi.org/10.48550/arXiv.1908.07442 (2020)

    2020 doi arXiv

  9. [18]

    & Friederich, P

    Ruff, R., Reiser, P., Stühmer, J. & Friederich, P. Connectivity Optimized Nested Graph Networks for Crystal Structures. Preprint at http://arxiv.org/abs/2302.14102 (2023)

    2023 arXiv

  10. [19]

    Ko, T. W. et al. Materials Graph Library (MatGL), an open-source graph deep learning library for materials science and chemistry. Preprint at https://doi.org/10.48550/arXiv.2503.03837 (2025)

    2025 doi arXiv

  11. [20]

    Goodall, R. E. A. & Lee, A. A. Predicting materials properties without crystal structure: deep representation learning from stoichiometry. Nat Commun 11, (2020)

    2020

  12. [21]

    Rhodes, B. et al. Orb-v3: atomistic simulation at scale. Preprint at https://doi.org/10.48550/arXiv.2504.06231 (2025)

    2025 doi arXiv

  13. [22]

    & Mizukami, W

    Shiota, T., Ishihara, K. & Mizukami, W. Universal neural network potentials as descriptors: towards scalable chemical property prediction using quantum and classical computers. Digital Discovery 3, 1714–1728 (2024)

    2024

  14. [23]

    El-Samman, A. M. et al. Global geometry of chemical graph neural network representations in terms of chemical moieties. Digital Discovery 3, 544–557 (2024)

    2024

  15. [24]

    M., De Castro, S., Morton, B

    El-Samman, A. M., De Castro, S., Morton, B. & De Baerdemacker, S. Transfe r learning graph representations of molecules for pKa,13 C-NMR, and solubility. Can. J. Chem. 102, 275–288 (2024)

    2024

  16. [25]

    Elijošius, R. et al. Zero shot molecular generation via similarity kernels. Nat Commun 16, 5991 (2025)

    2025

  17. [26]

    Y., Park, Y

    Kim, S. Y., Park, Y. J. & Li, J. Leveraging neural network inte ratomic potentials for a foundation model of chemistry. Preprint at https://doi.org/10.48550/arXiv.2506.18497 (2025)

    2025 doi arXiv

  18. [27]

    L., Buonassisi, T

    Oviedo, F., Ferres, J. L., Buonassisi, T. & Butler, K. T. Interpretable a nd Explainable Machine Learning for Materials Science and Chemistry. Acc. Mater. Res. 3, 597–607 (2022)

    2022

  19. [28]

    Machine learning in materials science: From explai nable predictions to autonomous design

    Pilania, G. Machine learning in materials science: From explai nable predictions to autonomous design. Computational Materials Science 193, (2021)

    2021

  20. [29]

    & Kumacheva, E

    Abolhasani, M. & Kumacheva, E. The rise of self-driving labs in chemi cal and materials sciences. Nat. Synth 2, 483–492 (2023)

    2023

  21. [31]

    De Breuck, P.-P., Evans, M. L. & Rignanese, G.-M. Robust model benchmarking and bias- imbalance in data-driven materials science: a case study on MODN et. J. Phys.: Condens. Matter 33, 404002 (2021)

    2021

  22. [33]

    & Tavazza, F

    Choudhary, K., Kalish, I., Beams, R. & Tavazza, F. High-throughput Identification a nd Characterization of Two-dimensional Materials using Density functional t heory. Sci Rep 7, 5179 (2017)

    2017

  23. [37]

    Jain, A. et al. Commentary: The Materials Project: A materials genome approach to accelerating materials innovation. APL materials 1, 11002 (2013)

    2013

  24. [38]

    & Aihara Jr., T

    Kawazoe, Y., Masumoto, T., Tsai, A.-P., Yu, J.-Z. & Aihara Jr., T. 1 Introduc tion. in Nonequilibrium Phase Diagrams of Ternary Amorphous Alloys (eds. Kawazoe, Y., Yu, J.- Z., Tsai, A.-P. & Masumoto, T.) vol. 37A 1–32 (Springer-Verlag, Berlin/Heidelberg, 1997)

    1997

  25. [39]

    & Wolverton, C

    Ward, L., Agrawal, A., Choudhary, A. & Wolverton, C. A general-purpose machine learning framework for predicting properties of inorganic materials. npj Comput Mater 2, 16028 (2016)

    2016

  26. [40]

    De Jong, M. et al. Charting the complete elastic properties of inorganic crystalline compounds. Sci Data 2, 150009 (2015)

    2015

  27. [42]

    Shen, J. et al. Reflections on one million compounds in the open quantum materials database (OQMD). J. Phys. Mater. 5, 031001 (2022)

    2022

  28. [43]

    Schmidt, J. et al. Machine‐Learning‐Assisted Determination of the Global Zero‐ Temperature Phase Diagram of Materials. Advanced Materials 35, 2210788 (2023)

    2023

  29. [44]

    Barroso-Luque, L. et al. Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models. Preprint at https://doi.org/10.48550/arXiv.2410.12771 (2024)

    2024 doi arXiv

  30. [45]

    Materialsproject.org https://matbench.materialsproject.org/Full%20Benchmark%20Data/matbench_v0.1_Meg Net_kgcnn_v2.1.0/ (2020)

    matbench_v0.1: MegNet (kgcnn v2.1.0) - MatBench. Materialsproject.org https://matbench.materialsproject.org/Full%20Benchmark%20Data/matbench_v0.1_Meg Net_kgcnn_v2.1.0/ (2020)

    2020

  31. [46]

    & Qian, Q

    Xu, Y. & Qian, Q. i-SISSO: Mutual information-based improved sure indepe ndent screening and sparsifying operator algorithm. Engineering Applications of Artificial Intelligence 116, 105442 (2022)

    2022

  32. [47]

    Jiang, X., Liu, G., Xie, J. & Hu, Z. Boosting SISSO Performance on Small Sample Datasets by Using Random Forests Prescreening for Complex Feature Selection. Prep rint at https://doi.org/10.48550/ARXIV.2409.19209 (2024)

    2024 doi arXiv

  33. [48]

    Foppa, L., Purcell, T. A. R., Levchenko, S. V., Scheffler, M. & Ghiringhelli, L. M. Hierarchical Symbolic Regression for Identifying Key Physical Parame ters Correlated with Bulk Properties of Perovskites. Phys. Rev. Lett. 129, 055301 (2022)

    2022

  34. [49]

    & Guestrin, C

    Chen, T. & Guestrin, C. XGBoost: A Scalable Tree Boosting System. in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 785–794 (ACM, San Francisco California USA, 2016). doi:10.1145/2939672.2939785

    2016

  35. [50]

    & Ong, S

    Chen, C., Ye, W., Zuo, Y., Zheng, C. & Ong, S. P. Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals. Chem. Mater. 31, 3564–3572 (2019)

    2019

  36. [51]

    Purcell, T. A. R., Scheffler, M. & Ghiringhelli, L. M. Recent advance s in the SISSO method and their implementation in the SISSO++ code. The Journal of Chemical Physics 159, 114110 (2023). Supplementary Information - Combining feature-based approaches with graph neural networ...

    2023

  37. [52]

    TMetalFraction|transition metal fraction

    and allow faster featurization. For the OFM features, the GNN-derived latent representation performs nearly as well as the original, with only a 0.0051 eV decline. Comb ining latent features from both GNN models slightly reduces the MAE, highlighting the potent ial benefits of...

  38. [53]

    Lam Pham, T. et al. Machine learning reveals orbital interaction in materials. Science and Technology of Advanced Materials 18, 756–765 (2017)

    2017

  39. [54]

    & Ong, S

    Chen, C., Ye, W., Zuo, Y., Zheng, C. & Ong, S. P. Graph Networks as a Universa l Machine Learning Framework for Molecules and Crystals. Chem. Mater. 31, 3564–3572 (2019)

    2019

  40. [55]

    & Jain, A

    Dunn, A., Wang, Q., Ganose, A., Dopp, D. & Jain, A. Benchmarking materials prope rty prediction methods: the Matbench test set and Automatminer reference algorithm. npj Computational Materials 6, 1–10 (2020)

    2020

  41. [56]

    Chollet, F. Keras. (2015)

    2015

  42. [57]

    & Ghiringhelli, L

    Ouyang, R., Curtarolo, S., Ahmetcik, E., Scheffler, M. & Ghiringhelli, L. M. SISSO: A compressed- sensing method for identifying the best low-dimensional descriptor in an immens ity of offered candidates. Phys. Rev. Materials 2, 083802 (2018)

    2018

  43. [58]

    Mechanical properties of some steels

    Citrine Informatics. Mechanical properties of some steels

  44. [59]

    & Tavazza, F

    Choudhary, K., Kalish, I., Beams, R. & Tavazza, F. High-throughput Identification a nd Characterization of Two-dimensional Materials using Density functional theory. Sci Rep 7, 5179 (2017)

    2017

  45. [60]

    Petretto, G. et al. High-throughput density-functional perturbation theory phonons for inorganic materials. Sci Data 5, 180065 (2018)

    2018

  46. [61]

    & Brgoch, J

    Zhuo, Y., Mansouri Tehrani, A. & Brgoch, J. Predicting the Band Gaps of Inorganic Solids by Machine Learning. J. Phys. Chem. Lett. 9, 1668–1673 (2018)

    2018

  47. [62]

    & Aihara Jr., T

    Kawazoe, Y., Masumoto, T., Tsai, A.-P., Yu, J.-Z. & Aihara Jr., T. 1 Introduction. in Nonequilibrium Phase Diagrams of Ternary Amorphous Alloys (eds. Kawazoe, Y., Yu, J.-Z., Tsai, A.-P. & Masumoto, T.) vol. 37A 1–32 (Springer-Verlag, Berlin/Heidelberg, 1997)

    1997

  48. [63]

    & Wolverton, C

    Ward, L., Agrawal, A., Choudhary, A. & Wolverton, C. A general-purpose machine le arning framework for predicting properties of inorganic materials. npj Comput Mater 2, 16028 (2016)

    2016

  49. [64]

    Petousis, I. et al. High-throughput screening of inorganic compounds for the discovery of novel dielectric and optical materials. Sci Data 4, 160134 (2017)

    2017

  50. [65]

    Castelli, I. E. et al. New cubic perovskites for one- and two-photon water splitting using the computational materials repository. Energy Environ. Sci. 5, 9034 (2012)

    2012

  51. [66]

    De Jong, M. et al. Charting the complete elastic properties of inorganic crystalline compounds . Sci Data 2, 150009 (2015)

    2015

  52. [67]

    Jain, A. et al. Commentary: The Materials Project: A materials genome approach t o accelerating materials innovation. APL materials 1, 11002 (2013)

    2013

  53. [68]

    Trinquet, V. et al. Second-harmonic generation tensors from high-throughput density-functional perturbation theory. Sci Data 11, 757 (2024)

    2024

  54. [69]

    Lundberg, S. M. & Lee, S.-I. A unified approach to interpreting model predi ctions. in Advances in neural information processing systems (eds. Guyon, I. et al.) vol. 30 (Curran Associates, Inc., 2017)

    2017

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.