Pith. sign in

REVIEW 5 major objections 8 minor 29 references

A machine-learning pipeline designs MOF photocatalysts predicted to beat existing materials by 125 percent.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

An ML pipeline combining reinforcement learning and graph neural networks claims to design MOF photocatalysts that outperform literature controls by over 125% in predicted fitness.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection Ambitious, well-organized ML pipeline for MOF design, but the headline performance claims rest on a closed evaluation loop and an uncalibrated fitness function; the workflow is worth a referee's time, the numbers are not. the 5 major comments →

arxiv 2607.27368 v1 pith:2F62OSI6 submitted 2026-07-29 cond-mat.mtrl-sci

Machine Learning for Designing Undesignable Metal-Organic Frameworks

Satya Kokonda This is my paper
classification cond-mat.mtrl-sci
keywords metal-organic frameworksphotocatalysisCO2 reductionreinforcement learninggraph neural networksmulti-objective optimizationmaterials designfitness function
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 materials considered too complex to design computationally—such as photocatalysts for CO2 reduction—can be designed by an automated pipeline that generates millions of candidate metal-organic frameworks, predicts their properties with graph neural networks, and scores them with a multi-objective fitness function. The workflow produced 20 MOFs predicted to outperform a control photocatalyst by over 125%, including a Zn-based MOF that beats the best control on every measured metric and a Cr-based MOF scoring 230% higher. If the approach holds, it would turn 'undesignable' applications into tractable design problems, extending beyond photocatalysis to drug delivery and energy storage. The paper's central claim is that the fitness function and funnel-based filtering together capture enough real-world chemistry to guide discovery of industrially relevant materials.

Core claim

The paper's central claim is that a reinforcement-learning generator, coupled to a funnel of crystal-graph neural-network predictors and a multiplicative fitness function, can design metal-organic frameworks for applications too complex for direct simulation. Applied to photocatalytic CO2 reduction, the workflow generated 60,000 novel MOFs, filtered them through thirteen predicted property channels, and identified 20 candidates with predicted photocatalytic viability at least 125% higher than a literature control. The strongest specific result: a Zn-based MOF outperforms the best control material across all five evaluation categories—stability, catalytic activity, cost, adsorption, and susta

What carries the argument

The central mechanism is a multiplicative fitness function combining thirteen predicted material properties—band gap, Faradaic efficiency, free energy, voltage potential, bulk modulus, water stability, cost, sustainability, thermal stability, available surface area, CO2 heat of adsorption, CO2/H2O selectivity, and synthesizability. Unlike additive scoring, the multiplicative form severely penalizes any single weak feature, reducing the chance that a candidate wins by excelling on one property while failing on essential others. This fitness function is fed by a funnel of thirteen graph-neural-network predictors that iteratively remove the lowest-scoring 5% of candidates at each stage, improvi

Load-bearing premise

The load-bearing premise is that the hand-constructed fitness function—combining predicted band gap, Faradaic efficiency, free energy, stability, cost, sustainability, and adsorption terms with literature constants—is a valid proxy for real photocatalytic CO2-reduction performance, since the controls are scored by the same unvalidated formula.

What would settle it

Synthesize the Zn-based MOF and the top Cr-based MOF and measure their actual photocatalytic CO2-to-CO conversion rates under standard solar illumination; if either performs within the range of the control materials rather than exceeding them by more than 125%, the central claim collapses. A less expensive check: compute the fitness function for a library of experimentally characterized MOF photocatalysts and test whether the fitness ranking correlates with measured activity.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the fitness function faithfully represents photocatalytic performance, the workflow can identify MOFs that outperform known photocatalysts by more than double, providing concrete experimental targets for CO2-to-CO conversion.
  • The multiplicative fitness design provides a template for other multi-objective materials problems, because it demands simultaneous competence across all design criteria rather than narrow optimization.
  • The recurrence of a specific metal cluster (N262) in top candidates suggests that the pipeline can extract chemical design rules—such as preferring waste-derived metals and semi-open topologies—that can guide future screening.
  • The reported 276% computational-efficiency gain from the funnel system could make trillion-scale MOF screening feasible where exhaustive simulation would be intractable.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A natural testable extension is to compare the fitness function's rankings against experimentally measured photocatalytic activities for a set of known MOFs; if the correlation is weak, the 125% improvement is an artifact of the scoring proxy rather than a real performance gain.
  • The paper's claim that the workflow is 'generalizable' to drug delivery and energy storage is plausible but unsubstantiated; those applications require different property channels and validation protocols, so the generalization is an extrapolation.
  • The 125% and 230% figures compare generated MOFs against only seven literature controls; a broader control set, including MOFs from a diverse experimental database, would likely reduce these margins but could still show meaningful gains.
  • Because the fitness function includes cost and sustainability terms derived from manually curated datasets, the practical upside of the approach may depend more on the quality of those datasets than on the ML architecture itself.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 8 minor

Summary. The manuscript proposes a machine-learning pipeline for the inverse design of metal-organic frameworks (MOFs) for photocatalytic CO2 reduction. A modified MOFReinforce model generates 60,000 MOFs optimized for predicted CO2/H2O selectivity. A funnel of CGCNN predictors filters these candidates on stability, catalytic descriptors, cost, sustainability, thermal stability, adsorption, and synthesizability, reducing the set to 10,986 candidates. A hand-constructed multiplicative fitness function (Eq. 4) ranks the survivors, and the authors claim that two MOFs—one Cr-based and one Zn-based—outperform literature controls by large margins, with 20 generated MOFs showing over 125% improvement over the best control. XRD patterns are simulated for the two finalists, and design patterns (e.g., metal cluster N262, topology bcg) are discussed. The central claims rest entirely on ML-predicted properties; no experimental or DFT validation is provided. Section IV.C explicitly acknowledges that the ML models 'may not always align with physical reality.'

Significance. If the results were supported by independent validation, the workflow would be a valuable general methodology for multi-objective material design in applications where direct simulation is impractical. The paper has several commendable features: the code is provided at a public repository, the funnel architecture is transparent, and the incorporation of cost, sustainability, and synthesizability criteria goes beyond single-property optimization. The paper also discloses some limitations in Section IV.C, including small training sets and reliance on ML predictions. However, the quantitative claims of 125–230% improvements are not established. The evaluation loop is closed for two key descriptors (selectivity and heat of adsorption), the fitness function is not calibrated to any experimental or DFT outcome, and the final candidates are not validated. As presented, the results demonstrate an internally consistent screening algorithm but not the design of industrially promising photocatalysts.

major comments (5)
  1. [Section II.C, Figs. 1 and 3] Closed evaluation loop for selectivity and heat of adsorption. The CGCNN that scores the generated MOFs is trained on 'MOFReinforce predictor information converted to CIF files'—i.e., labels produced by the same predictor that served as the RL reward during generation. Literature controls were not generated by MOFReinforce and are therefore out-of-distribution for this scorer. Any systematic bias in the MOFReinforce predictor is inherited by the scoring model, so the claimed >125% improvements in Fig. 7 and the radar-chart advantages in Fig. 5 may reflect score inflation rather than physical superiority. Section IV.C concedes ML predictions 'may not always align with physical reality' but does not address this specific distribution shift. Independent DFT or experimental labels for selectivity and heat of adsorption are needed before these comparative claims can be trusted.
  2. [Section II.D, Eq. (4)] The fitness function is ad hoc and unvalidated as a proxy for photocatalytic performance. The coefficients and ideal ranges in Eqs. (2) and (3) are hand-selected, and Eq. (4) multiplies terms whose normalization and units are not specified. No calibration to measured photocatalytic CO2 reduction is reported. The control materials are scored with the same formula, so the '230% higher' and '125% better' statements are statements about this particular fitness function, not about real-world performance. The paper should either validate Eq. (4) against known photocatalysts, provide a sensitivity analysis, or clearly reframe all headline numbers as 'predicted fitness scores' rather than as improvements in photocatalytic viability.
  3. [Section II.C and Section IV.C] The water-stability filter is load-bearing but is trained on approximately 36 samples, as the paper itself notes in the limitations bullet ('only 36 samples'). This model is used to remove the bottom 5% of 19,012 MOFs, i.e., roughly 950 candidates. With n=36 and no reported accuracy or uncertainty, such a hard cutoff cannot be justified. This undermines the reliability of the funnel and, in turn, the list of final candidates.
  4. [Section II.C, Eq. (3)] The stated intent for the heat-of-adsorption function is to 'incorporate viable materials found with HOA near -60 kJ/mol' via a slow taper. However, with f(x)=exp(-(x+30)^2/150), H(-60) ≈ 0.005, so such materials are essentially excluded, not tapered in. This contradiction directly affects the adsorption component of Eq. (4) and should be corrected or justified.
  5. [Section III, Figs. 5–7; Section V] The benchmark comparison is underspecified. Only one control material (PCN-224-Zr) is named; the other six literature controls are not listed, their selection criteria are not given, and no raw fitness scores or uncertainties are reported. The headline claim that 20 generated MOFs show over 125% improvement is based on predicted scores against a single 'best control.' The abstract and conclusion go further and describe the materials as 'industrially promising' and 'designed,' despite the absence of experimental or DFT validation. These claims exceed what the evidence can support.
minor comments (8)
  1. [Abstract] The abstract contains a typo: 'CO/H20 selectivity' should be 'CO2/H2O selectivity.'
  2. [Section II.B, Eq. (1)] Equation (1) is typeset ambiguously: 'FaradaicEfficiency 5 −5·FreeEnergy− |VP| 2' is hard to parse. Define the division clearly (Fara/5 and |VP|/2).
  3. [Section II.C] The claimed '276% improvement in computational efficiency' and the later '6.54×10^7 times speedup' in Section IV.B are not derived from a stated baseline. Please define the reference method and the formula used for the speedup.
  4. [Section II.B] The 18% reduction in MAE and 36% improvement in computational efficiency for the CGCNN modification are stated without baseline details or evaluation datasets. Provide the comparison protocol.
  5. [Eq. (2) vs. Eq. (4)] Equation (2) defines B(x), but Eq. (4) uses 'BandGap' in the catalytic bracket. Clarify whether the raw band gap or the transformed B(x) enters the fitness function.
  6. [Section III, Fig. 4] The metal cluster identifier 'N262' is not chemically defined. Provide the composition of this cluster or a reference figure so the reader can interpret the design-pattern analysis.
  7. [Section III, Fig. 6] The similarity of simulated XRD patterns to QMOF patterns is presented as evidence of viability, but XRD similarity is not a sufficient synthesis criterion. This should be stated explicitly in the text.
  8. [References] Reference [8] is a commercial web page rather than a peer-reviewed source; please cite primary literature for the biomedical applications of CO.

Circularity Check

2 steps flagged

Closed loop: selectivity/HOA scorer is trained on the same predictor used as RL reward, so the reported >125% improvement is partly a self-consistency check, not independent prediction.

specific steps
  1. fitted input called prediction [Section II.C (Funnel System); Fig. 1 caption; Eq. 4; Fig. 7]
    "The generated MOFs are tokenized and passed through a neural network to estimate CO2/H2O selectivity. These predictions are then used as feedback to optimize the generator via reinforcement learning... A database used for training the CGCNN to predict CO2 Heat of Adsorption (HOA) and CO2/H2O selectivity was created from the MOFReinforce predictor information converted to CIF files."

    CO2/H2O selectivity is both the RL reward for the generator and a term in the final fitness function (Eq. 4). The CGCNN that later scores the generated MOFs and the literature controls was trained on labels produced by that same MOFReinforce selectivity predictor. Thus the selector and scorer are the same model, just re-encoded; the generated MOFs were optimized to satisfy it while the controls were not. The reported '125% improvement in predicted photocatalytic viability' (Fig. 7) and the adsorption/selectivity advantages (Fig. 5) therefore partly measure how well the generator overfits its own reward model, not an independent property of the materials.

  2. self definitional [Section II.D (Eq. 4); Section III (Fig. 7); Section IV.C]
    "This designed function provides a computational method to model photocatalytic performance... 20 candidates demonstrated over 125% improvement in predicted photocatalytic viability."

    Photocatalytic 'viability' is defined by the hand-constructed Eq. 4, and the top 20 candidates are selected by ranking that same function; the literature controls are then scored with the same function. The 125% figure is therefore a difference in the optimization objective itself, not an externally anchored measurement. The paper's own limitation statement—'the model's performance is tied to a fitness function that combines multiple feature scores into a single optimization objective'—concedes that the headline improvement is a property of the chosen metric. Without independent calibration of Eq. 4 to experimental photocatalytic outcomes, the claim reduces to 'our selected candidates score higher on our score.'

full rationale

The central reported improvements are not independent of the optimization target: the selectivity/HOA CGCNN is trained on labels from the same MOFReinforce predictor used as RL reward, and the final fitness function defines the 'photocatalytic viability' being predicted. This makes the >125% figure partly a closed-loop self-consistency result. The rest of the pipeline (UFF geometry, synthesis filters, XRD similarity, cost/sustainability data) is independent and non-circular; the paper also includes a limitations section admitting model accuracy may not align with physical reality. However, because the headline quantitative claim is obtained by optimizing and then evaluating with the same unvalidated metric, the circularity is more than minor.

Axiom & Free-Parameter Ledger

7 free parameters · 6 axioms · 2 invented entities

The paper contributes a large number of hand-chosen constants and ad hoc thresholds in the funnel and fitness function, while relying on unvalidated assumptions that ML-predicted descriptors and a closed training loop can substitute for experimental performance. The two designed MOFs are new but have no independent experimental evidence.

free parameters (7)
  • Band gap bell-curve amplitude = 4.536
    Constant in Eq. 2 chosen to normalize the bell curve; no physical derivation.
  • Band gap optimum center = 1.9 eV
    Center of B(x) from the stated ideal range 1.8-2 eV; chosen by hand.
  • Band gap curve width = 150
    Denominator in the exponent of Eqs. 2 and 3; ad hoc width controlling tolerance.
  • Heat-of-adsorption ideal center = -30 kJ/mol
    Center of g(x) in Eq. 3, midpoint of the approximated ideal range -20 to -40 kJ/mol; labeled 'approximated' in the text.
  • Fitness coefficients in Eq. 4 = e.g., Fara/5, FreeEnergy*5, |VP|/2
    Weights inherited from CarbNN Eq. 1 for electrocatalysis and applied without re-calibration to photocatalysis.
  • Funnel trim fractions = 15% first, then 5% at each stage
    Thresholds for iterative removal of low-scoring MOFs; chosen ad hoc, not optimized or justified.
  • Synthesis classification cutoff = 0.5
    CGCNN classification threshold to decide which MOFs remain; no stated basis for the value.
axioms (6)
  • domain assumption Electrocatalytic descriptors transfer to photocatalysis
    Section II.C: 'features shown to impact electrocatalytic success are extrapolated to photocatalytic success.' This is asserted, not demonstrated.
  • ad hoc to paper Fitness function Eq. 4 is a valid proxy for photocatalytic performance
    No calibration to experimental photocatalytic MOF data; constants and ranges set by hand; the only 'validation' is that selected materials score high on the same function.
  • domain assumption CGCNN predictions are sufficiently accurate for relative rankings
    Models are trained on QMOF/MOFSimplify/MOSAEC data; the water-stability model is admitted to use only 36 samples; no uncertainty quantification is provided.
  • ad hoc to paper MOFReinforce selectivity predictor can serve as ground truth for training CGCNN
    Section II.C uses 'MOFReinforce predictor information' to build training data for HOA and selectivity, making the evaluation model a proxy of the generator's own reward.
  • domain assumption Simulated XRD patterns indicate structural plausibility and synthesizability
    Section III compares simulated XRD (pymatgen) of designed MOFs to QMOF structures; no experimental XRD or synthesis is performed.
  • domain assumption UFF-optimized structures are realistic
    UFF geometry optimization (Section II.A) is standard but does not guarantee that generated hypothetical MOFs are experimentally realizable.
invented entities (2)
  • Cr-based MOF candidate no independent evidence
    purpose: Proposed photocatalyst for CO2 reduction; claimed 230% higher predicted photocatalyst score than control.
    Only ML-predicted fitness and simulated XRD; no experimental synthesis, characterization, or measured catalytic activity.
  • Zn-based MOF candidate no independent evidence
    purpose: Proposed photocatalyst claimed to outperform the best control across all five evaluated metrics.
    Same lack of experimental validation; ranking depends on the hand-built Eq. 4 fitness function and predicted properties.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Machine Learning for Designing Undesignable Metal-Organic Frameworks." pith.science (2026). https://pith.science/paper/2F62OSI6

@misc{pith2026260727368,
  author       = {Pith},
  title        = {Pith review of: Machine Learning for Designing Undesignable Metal-Organic Frameworks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2F62OSI6}},
  note         = {Machine review of arXiv:2607.27368}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Many crucial processes are too complex for computational modeling, requiring experimentation to identify promising materials. Here, a methodology for material design is presented, while photocatalysis is presented as a specific case-study. Metal-Organic Frameworks (MOFs) are a subset of highly promising porous nanomaterials, used in a variety of unmodellable applications. Reinforcement learning generated 60,000 novel MOFs optimized for CO/H20 selectivity. A predictor funnel system was created, iteratively removing low-scoring MOFs to 10,986 potential candidates, improving computational efficiency by 276%. While trained Crystal Graph Convolutional Neural Network (CGCNN) models predicted features for creating a fitness function incorporating stability, catalytic ability, material cost, sustainability, and adsorption while allowing the inclusion of application specific design criterion. This designed function provides a computational method to model photocatalytic performance- and filtered down to two promising MOFs which each pass a myriad of synthesis criteria, first a Cr-based MOF with photocatalyst score 230% higher than the control. Second, a Zn-based MOF outperforms the best control across all relevant metrics, demonstrating robustness against variable fitness functions. This work designed 20 materials, each 125% better than the control for this application. Furthermore, analysis revealed insightful design patterns, such as the significant influence of metal cluster N262 on catalytic performance, providing a method for future work to narrow the chemical space. By incorporating industrially applicable features such as cost or stability of the material, this work successfully designs industrially promising materials in otherwise unmodellable processes such as drug delivery, while paving a method for multi-objective optimization incorporating 260% more features than prior work.

Figures

Figures reproduced from arXiv: 2607.27368 by Satya Kokonda.

Figure 1
Figure 1. Figure 1: Diagram of the MOFReinforce encoder–decoder architecture. The [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Post-processing workflow applied following MOF generation. This [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: The distribution of metal clusters and organic linkers among the top [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 7
Figure 7. Figure 7: Performance comparison between the 20 highest-ranked generated [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗
Figure 6
Figure 6. Figure 6: X-ray diffraction (XRD) patterns of the two promising MOFs com [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

29 extracted references · 2 linked inside Pith

  1. [1]

    Ten chemical innovations that will change our world: Iupac identifies emerging technologies in chemistry with potential to make our planet more sustainable,

    F. Gomollon-Bel, “Ten chemical innovations that will change our world: Iupac identifies emerging technologies in chemistry with potential to make our planet more sustainable,”Chemistry International, vol. 41, no. 2, pp. 12–17, 2019

    2019

  2. [2]

    Inverse design of metal–organic frameworks for direct air capture of CO 2 via deep reinforcement learning,

    H. Park, S. Majumdar, X. Zhang, J. Kim, and B. Smit, “Inverse design of metal–organic frameworks for direct air capture of CO 2 via deep reinforcement learning,”Digital Discovery, vol. 3, no. 4, pp. 728–741, 2024

    2024

  3. [3]

    Metal-organic frameworks as electrolyte additives to enable ultrastable plating/stripping of Li anode with dendrite inhibition,

    F. Chu, J. Hu, C. Wu, Z. Yao, J. Tian, Z. Li, and C. Li, “Metal-organic frameworks as electrolyte additives to enable ultrastable plating/stripping of Li anode with dendrite inhibition,”ACS Applied Materials & Inter- faces, vol. 11, no. 4, pp. 3869–3879, 2019

    2019

  4. [4]

    PEG functionalized zirconium dicarboxylate MOFs for docetaxel drug delivery in vitro,

    V . Gupta, S. Mohiyuddin, A. Sachdev, P. Soni, P. Gopinath, and S. Tyagi, “PEG functionalized zirconium dicarboxylate MOFs for docetaxel drug delivery in vitro,”Journal of Drug Delivery Science and Technology, vol. 52, pp. 846–855, 2019

    2019

  5. [5]

    Loading of MOF-5 with Cu and ZnO nanoparticles by gas-phase infiltration with organometallic precursors: Properties of Cu/ZnO@MOF-5 as catalyst for methanol synthesis,

    M. M ¨uller, S. Hermes, K. K ¨ahler, M. W. E. van den Berg, M. Muhler, and R. A. Fischer, “Loading of MOF-5 with Cu and ZnO nanoparticles by gas-phase infiltration with organometallic precursors: Properties of Cu/ZnO@MOF-5 as catalyst for methanol synthesis,”Chemistry of Materials, vol. 20, no. 14, pp. 4576–4587, 2008

    2008

  6. [6]

    Benchmarking density functional theory predictions of framework structures and properties in a chemically diverse test set of metal–organic frameworks,

    D. Nazarian, P. Ganesh, and D. S. Sholl, “Benchmarking density functional theory predictions of framework structures and properties in a chemically diverse test set of metal–organic frameworks,”Journal of Materials Chemistry A, vol. 3, no. 44, pp. 22 432–22 440, 2015

    2015

  7. [7]

    Towards a MOF e-nose: A surmof sensor array for detection and discrimination of plant oil scents and their mixtures,

    S. Okur, Z. J. Zhang, M. Sarheed, P. Nick, U. Lemmer, and L. Heinke, “Towards a MOF e-nose: A surmof sensor array for detection and discrimination of plant oil scents and their mixtures,”Sensors and Actuators B: Chemical, vol. 306, p. 127502, 2020

    2020

  8. [8]

    Applications of carbon monoxide (CO) in information industry,

    T. Gao, “Applications of carbon monoxide (CO) in information industry,” https://gaotek.com/ applications-of-carbon-monoxide-co-in-information-industry/, 2023, accessed: 2023-11-27

    2023

  9. [9]

    High selectivity of photocatalytic reduction of CO2 to CO based on terpyridine ligand supported CuI metal organic framework,

    W. Zhang, Y . Wang, Y . Liang, A. Jiang, H. Gong, X.-Y . Tian, W. Fu, J. Liao, P. Chen, and Y . Ma, “High selectivity of photocatalytic reduction of CO2 to CO based on terpyridine ligand supported CuI metal organic framework,”Frontiers in Chemistry, vol. 10, 2022

    2022

  10. [10]

    Multi-scale computational design of metal– organic frameworks for carbon capture using machine learning and multi-objective optimization,

    Z. Deng and L. Sarkisov, “Multi-scale computational design of metal– organic frameworks for carbon capture using machine learning and multi-objective optimization,”Chemistry of Materials, 2024

    2024

  11. [11]

    CarbNN: A novel active transfer learning neural network to build de novo metal organic frameworks (MOFs) for carbon capture,

    N. Redkar, “CarbNN: A novel active transfer learning neural network to build de novo metal organic frameworks (MOFs) for carbon capture,” https://arxiv.org/abs/2311.16158, 2023

    Pith/arXiv arXiv 2023

  12. [12]

    Sustainable synthesis of metal-organic frameworks and their derived materials from organic and inorganic wastes,

    K. Boukayouht, L. Bazzi, and S. El Hankari, “Sustainable synthesis of metal-organic frameworks and their derived materials from organic and inorganic wastes,”Coordination Chemistry Reviews, vol. 478, p. 214986, 2023

    2023

  13. [13]

    POWER 101: Flue gas heat recovery in power plants, part I,

    POWER, “POWER 101: Flue gas heat recovery in power plants, part I,” https://www.powermag.com/ power-101-flue-gas-heat-recovery-in-power-plants-part-i/, 2010, accessed: 2025-04-24

    2010

  14. [14]

    Tuning the mechanical response of metal– organic frameworks by defect engineering,

    R. A. Fischer, P. Vervoorts, C. L. Hobday, T. D ¨uren, D. Daisenberger, A. Smith, and G. Kieslich, “Tuning the mechanical response of metal– organic frameworks by defect engineering,”Journal of the American Chemical Society, vol. 140, no. 37, pp. 11 581–11 584, 2018

    2018

  15. [15]

    MOFDiff: Coarse- grained diffusion for metal-organic framework design,

    X. Fu, T. Xie, A. S. Rosen, T. Jaakkola, and J. Smith, “MOFDiff: Coarse- grained diffusion for metal-organic framework design,” https://arxiv.org/ abs/2310.10732, 2023, arXiv preprint

    Pith/arXiv arXiv 2023

  16. [16]

    Process integration for the biodiesel production from biomitigation of flue gases,

    R. J. Barla, S. Raghuvanshi, and S. Gupta, “Process integration for the biodiesel production from biomitigation of flue gases,” inElsevier EBooks. Elsevier, 2022, pp. 191–215

    2022

  17. [17]

    UFF, a full periodic table force field for molecular mechanics and molecular dynamics simulations,

    A. K. Rapp ´e, C. J. Casewit, K. S. Colwell, W. A. Goddard, and W. M. Skiff, “UFF, a full periodic table force field for molecular mechanics and molecular dynamics simulations,”Journal of the American Chemical Society, vol. 114, no. 25, pp. 10 024–10 035, 1992

    1992

  18. [18]

    RDKit: Open-source cheminformatics,

    RDKit Project, “RDKit: Open-source cheminformatics,” https://www. rdkit.org, 2025, accessed: 2025-04-24

    2025

  19. [19]

    GitHub – satyak-0/mofcreationn-model-1,

    SatyaK-0, “GitHub – satyak-0/mofcreationn-model-1,” https://github.com/SatyaK-0/MOFCreatioNN-Model-1/tree/main, 2024, accessed: 2025-04-24

    2024

  20. [20]

    Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties,

    T. Xie and J. C. Grossman, “Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties,” Physical Review Letters, vol. 120, no. 14, 2018

    2018

  21. [21]

    MOFSimplify, machine learning models with extracted stability data of three thousand metal–organic frameworks,

    A. Nandy, G. Terrones, N. Arunachalam, C. Duan, D. W. Kastner, and H. J. Kulik, “MOFSimplify, machine learning models with extracted stability data of three thousand metal–organic frameworks,”Scientific Data, vol. 9, no. 1, 2022

    2022

  22. [22]

    CO 2 reduction: From the electrochemical to photochemical approach,

    J. Wu, Y . Huang, W. Ye, and Y . Li, “CO 2 reduction: From the electrochemical to photochemical approach,”Advanced Science, vol. 4, no. 11, p. 1700194, 2017

    2017

  23. [23]

    Machine learning the quantum- chemical properties of metal–organic frameworks for accelerated ma- terials discovery,

    A. S. Rosen, S. M. Iyer, D. Ray, Z. Yao, A. Aspuru-Guzik, L. Gagliardi, J. M. Notestein, and R. Q. Snurr, “Machine learning the quantum- chemical properties of metal–organic frameworks for accelerated ma- terials discovery,”Matter, vol. 4, no. 5, pp. 1578–1597, 2021

    2021

  24. [24]

    MOSAEC-DB: A comprehensive database of experimental metal–organic frameworks with verified chemical accuracy suitable for molecular simulations,

    M. Gibaldi, A. Kapeliukha, A. White, R. Luo, R. A. Mayo, J. Burner, and T. K. Woo, “MOSAEC-DB: A comprehensive database of experimental metal–organic frameworks with verified chemical accuracy suitable for molecular simulations,”Chemical Science, 2025

    2025

  25. [25]

    Intelligent selection of metal-organic framework arrays for methane sensing via genetic algorithms,

    J. A. Gustafson and C. E. Wilmer, “Intelligent selection of metal-organic framework arrays for methane sensing via genetic algorithms,”ACS Sensors, vol. 4, no. 6, pp. 1586–1593, 2019

    2019

  26. [26]

    Python materials genomics (pymatgen): A robust, open-source python library for materials analysis,

    S. P. Ong, W. D. Richards, A. Jain, G. Hautier, M. Kocher, S. Cholia, D. Gunter, V . L. Chevrier, K. A. Persson, and G. Ceder, “Python materials genomics (pymatgen): A robust, open-source python library for materials analysis,”Computational Materials Science, vol. 68, pp. 314–319, 2013

    2013

  27. [27]

    Computational screening of trillions of metal–organic frameworks for high-performance methane storage,

    S. Lee, B. Kim, H. Cho, H. Lee, S. Y . Lee, E. S. Cho, and J. Kim, “Computational screening of trillions of metal–organic frameworks for high-performance methane storage,”ACS Applied Materials & Inter- faces, vol. 13, no. 20, pp. 23 647–23 654, 2021

    2021

  28. [28]

    Computational modeling guided design of metal–organic frameworks for photocatalysis – a mini review,

    S. Yu, Q. Wang, J. Wang, C. Fang, Y . Li, J. Ge, and B. Fang, “Computational modeling guided design of metal–organic frameworks for photocatalysis – a mini review,”Catalysis Science & Technology, vol. 13, no. 23, pp. 6583–6603, 2023

    2023

  29. [29]

    Current update and prospects in the development of conductive metal-organic framework electrodes for lithium-based batteries,

    S. O. Ajayi, T. H. Dolla, L. L. Sikeyi, A. O. Akinola, W. K. Maboya, X. Liu, P. R. Makgwane, and M. K. Mathe, “Current update and prospects in the development of conductive metal-organic framework electrodes for lithium-based batteries,”Materials Today Sustainability, vol. 27, p. 100899, 2024

    2024

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.