MOFGPT: Generative Design of Metal-Organic Frameworks using Language Models

Akshay Antony; Amir Barati Farimani; Radheesh Sharma Meda; Rishikesh Magar; Srivathsan Badrinarayanan

REVIEW 4 major objections 3 minor 90 references

MOFGPT: Generative Design of Metal-Organic Frameworks using Language Models

T0 review · 4 major / 3 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A GPT-style model trained on MOFid strings and tuned by reinforcement learning can generate valid, novel MOFs whose predicted gas-uptake and band-gap values land near chosen targets; supervised fine-tuning alone yields none.

desk verdict A plausible GPT+RL pipeline over MOFid strings, but the property-targeting claims are unverified because the same surrogate supplies both the reward and the reported evaluation. read the letter →

arxiv 2506.00198 v1 pith:KB6EPFIY submitted 2025-05-30 cs.LG cond-mat.mtrl-scics.AI

Srivathsan Badrinarayanan , Rishikesh Magar , Akshay Antony , Radheesh Sharma Meda , Amir Barati Farimani This is my paper

classification cs.LGcond-mat.mtrl-scics.AI MSC 68T07

keywords metal-organicframeworksgenerativedesignlanguagemodelsreinforcementlearningMOFidproperty-targetedgenerationinversetransformer

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

Metal-organic frameworks (MOFs) are crystals of metal nodes and organic linkers whose design space is too large to screen exhaustively with molecular simulation. This paper claims that MOF design can be converted into a text-generation problem: a GPT-style language model is trained on MOFid, a single-line encoding of a MOF's building blocks and topology, and then tuned by reinforcement learning with a reward that combines chemical validity, novelty, diversity, and closeness to a target property. The reported result is that supervised fine-tuning on property labels produces zero valid structures, whereas the RL-tuned model achieves validity rates from about 24% to 100% across eleven property tasks and shifts the predicted distributions of methane uptake, CO2 uptake, and band gap toward the chosen targets. If this holds up, generating targeted MOF candidates could become cheap enough to complement expensive simulation-based screening. The paper measures success with the same frozen property predictor that acts as the reward, which it acknowledges sets the ceiling on how well the optimization can track true physics.

What carries the argument

The load-bearing object is the MOFid string, which turns a crystal into a sentence: SMILES for the organic and inorganic secondary building units, a separator token, and an RCSR topology code with catenation. A GPT-2-style decoder with 12 layers and 768-dimensional embeddings generates these strings autoregressively, and a frozen transformer-based predictor supplies the property value used in the reward. The steering comes from the multi-component reward function $R = \beta_{\text{target}}R_{\text{target}} + \alpha_n R_{\text{novelty}} + \alpha_v R_{\text{validity}} + \alpha_d R_{\text{diversity}}$, where the target term is a tiered proximity reward with direction bonuses for higher-is-better versus lower-is-better objectives, the validity term enforces RDKit syntax checks plus metal-node, component-balance, topology, and coordination-number checks, and the diversity term blends batch, n-gram, history, and composition measures. The policy is updated with REINFORCE using a mean-reward baseline, protected by a KL term that keeps the policy near the pretrained model, and guided by a global memory of the 200 best discovered structures, an adaptive top-K selection that narrows from 50% to 30% of the batch as training proceeds, and conditional reward normalization.

What would settle it

Take the highest-scoring generated structures from Table 3, for example the nbo-topology Zn dicarboxylate MOF predicted at 2.414 mol/kg methane uptake and the pcu-topology Cu paddlewheel MOF predicted at 3.33 mol/kg CO2 uptake, reconstruct them as three-dimensional crystals, and compute adsorption with grand-canonical Monte Carlo simulation or band gaps with DFT; if the high-fidelity values deviate strongly from the predictor's outputs, or if the strings cannot be realized as physical crystals at all, the central claim that reinforcement learning drives genuine property improvement would be refuted. A cheaper control is to run the same reinforcement-learning procedure with a predictor whose labels are shuffled, and check whether the generated property distribution still moves toward the target.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that reward-guided sequence generation resolves a failure mode that supervised fine-tuning cannot. A GPT-2-style model pretrained on 323,469 MOFid strings learns the grammar of MOF chemistry; fine-tuning that model as a property regressor destroys the grammar, and the fine-tuned generator yields no chemically valid MOFs for any property target. Tuning the same pretrained model with REINFORCE, using a frozen transformer-based predictor as the value network and a four-term reward for validity, novelty, diversity, and tiered target proximity, restores validity while steering generation: across methane adsorption at five pressures, CO2 adsorption at five pressures, and electronic band gap, the model produces MOFs that pass syntax, component, topology, and coordination checks at rates between 23.94% and 100%, with novelty at least 63% and diversity at least 83%, and the mean predicted property of the generated sets lands near each requested target chosen from the mean and one or two standard deviations of the training distribution. The paper interprets the shifted distributions and the chemistry of the generated examples, such as Zn nodes with carboxylate linkers for methane uptake, Cu paddlewheel units with nitrogen- and oxygen-rich linkers for CO2 uptake, and conjugated donor-acceptor motifs for low band gaps, as evidence that the model learned structure-property relationships rather than memorized training structures.

Load-bearing premise

The load-bearing assumption is that the frozen machine-learning property predictor is an honest stand-in for the real physics, so that maximizing its output genuinely improves a MOF's gas uptake or band gap, an assumption the paper itself flags in its Limitations section, where it concedes the reinforcement-learning optimization is only as good as the property prediction model, and which no simulation or experiment in the paper verifies.

Editorial extensions

If this is right

Supervised fine-tuning on property labels is not a viable route to valid MOF generation; property-targeted generation requires reward-guided optimization that preserves the pretrained grammar.
Property distributions of generated MOFs can be pushed into the tail of the training distribution, roughly the top 4.5% of performances at the mean plus two standard deviations, without collapsing validity.
The same pipeline transfers across gas adsorption and electronic band-gap properties without domain-specific changes, so multi-property objectives are a plausible extension within the same framework.
Generated candidates encode chemically interpretable motifs, such as carboxylate linkers with Zn nodes in nbo topology for methane, Cu paddlewheels with nitrogen- and oxygen-rich linkers for CO2, and conjugated donor-acceptor units for low band gaps, consistent with known high-performing MOF chemistry.

Reading between the lines

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

The property improvements are predicted rather than confirmed: both the reward and the reported statistics come from the same frozen predictor, so the natural next test is to rebuild the top generated MOFids as three-dimensional crystals and run grand-canonical Monte Carlo adsorption or DFT band-gap calculations on them.
A control experiment the paper does not run would be reinforcement learning with a deliberately scrambled property predictor; if the generated property distribution still moves toward the target, then the reward signal is not what is causing the shift.
Because MOFid omits three-dimensional coordinates and the paper concedes that a separate structure-generation step adds uncertainty, the realized gain in real materials could be smaller than the reported distribution shifts suggest.
The framework's ceiling is the predictor's accuracy, which suggests an active-learning loop that retrains the predictor on high-fidelity simulation output for the most promising generated candidates would be the decisive testbed for whether the targeted shifts survive contact with physics.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Two things to know. First, the engineering is real: a GPT-2 decoder trained on MOFid strings, with a multi-component RL reward balancing validity, novelty, diversity, and property targeting, and a code/data link. To my knowledge that specific combination isn't in the cited prior work. Second, the central claim—that this drives the model toward MOFs with desired functional attributes—is not supported by independent evidence. The property predictor MOFormer provides the RL reward in Eq. (7) and also produces the property statistics in Table 2. That's a closed loop. The distribution shifts show the RL policy moved its own surrogate, not that real adsorption or band gaps changed. The SI Limitations (S14) concede as much: the optimization is only as good as the property prediction model. No GCMC, DFT, or experimental check appears anywhere.

The paper does several things well. It's clearly written, the reward design is detailed, and the appendix shows systematic multi-pressure experiments. The SI's explicit acknowledgment of the surrogate limitation is honest. And the authors did not bury the fact that some targets are hard; validity rates fluctuate a lot across conditions.

The soft spots, in order of weight. The evaluative circularity is the load-bearing one, and it's not fixable by adding more RL rows; it needs at least one physics-based validation pass on a sample of generated structures. The "fine-tuned models fail" result is also a strawman of sorts: the fine-tuned model is a property-prediction head, not a generative model fine-tuned on MOFid sequences, so "supervised fine-tuning fails to generate valid MOFs" is comparing different tasks. There are also internal inconsistencies that a careful referee will notice: the pressure lists in the main text (1.0 and 5.0 bar for CH4; 0.01, 0.1, 0.5, 1.0, 5.0 for CO2) don't match the SI (0.05, 0.5, 0.9, 2.5, 4.5; and 0.01, 0.05, 0.1, 0.5, 2.5). The results section says "three distinct tasks" then lists four. And several Table 2 rows don't move toward target—CO2 at 0.1 bar, mean+2σ, gets 0.432 vs original 0.599 and a target near 1.96; band gap mean+1σ lands at 1.767 instead of ~2.97. These aren't fatal to the engineering, but they weaken the "systematic targeting" narrative. No error bars, and the generation denominator shifts between descriptions.

Bottom line: I would not accept the property-targeting claims as established, so I wouldn't cite them. But I would send this to peer review, not desk reject. The pipeline is a reasonable contribution to a crowded but active area, and the circularity is fixable—add GCMC or DFT validation on a sample, replace the degenerate baseline with a real generative fine-tuned GPT-2 on MOFids, and reconcile the inconsistencies. That would be a solid paper. Right now it's a promising preprint that overstates its evidence.

Referee Report

4 major / 3 minor

Summary. The manuscript presents MOFGPT, a GPT-2-based generative model for Metal-Organic Frameworks represented as MOFid strings, trained in three stages: pretraining on a large corpus of MOFid sequences, supervised fine-tuning for property prediction, and REINFORCE-based reinforcement learning with a multi-component reward function (validity, novelty, diversity, and target property proximity) that uses a frozen MOFormer property predictor. The authors report that supervised fine-tuning produces 0% valid MOFs, whereas RL achieves validity rates ranging from roughly 16% to 100% across tasks, with high novelty and diversity, and shifts in predicted property distributions toward chosen targets for CH4 adsorption, CO2 adsorption, and band gap. The central claim is that this pipeline drives the model toward synthesizable, topologically valid MOFs with desired functional attributes, thus enabling target-specific inverse design of MOFs.

Significance. The idea of applying RL to MOFid sequence generation, with reward terms balancing validity, novelty, diversity, and property targeting, is timely and potentially useful for computational MOF discovery. Strengths include the large curated pretraining corpus (323,469 training and 81,260 test MOFids), a transparent multi-component reward specification (Eqs. 6-10 and SI Tables S4-S5), and validation/novelty/diversity reporting over many pressure conditions. However, as presented, the central claim rests on an unverified premise: that the MOFormer surrogate is accurate enough on novel generated structures that maximizing its outputs corresponds to genuinely improved physical properties. The evaluation loop is closed because the same surrogate provides both the RL reward and the reported success statistics, and no independent GCMC, DFT, or experimental validation is supplied. Some of the reported statistics also undercut the targeting claim (e.g., CO2 at 0.1 bar in Table 2). If the property signal were independently validated, this would be a valuable contribution; with the current evidence, the significance claim is not established.

major comments (4)

[Stage 3: Reinforcement Learning Framework; Eq. (7); Table 2; SI Limitations] The property-targeting evaluation is a closed loop: Rtarget in Eq. (7) is a function of p_hat(m), the MOFormer-predicted property value, and Table 2 reports exactly those same MOFormer-predicted values for the generated set. The distribution shifts in Figure 3 and Table 2 therefore demonstrate only that RL optimized the surrogate objective; they do not establish that the generated MOFs have genuinely improved adsorption capacities or band gaps. The SI Limitations section explicitly concedes that 'the reinforcement learning optimization is only as good as the underlying property prediction model,' and no GCMC, DFT, or experimental measurement is reported for any generated structure. To support the abstract's claim of 'desired functional attributes,' the authors should reconstruct 3D structures from the generated MOFids and evaluate a representative subset with an independent method (GCMC for adsorption, DFT for band gaps), reporting agreement between predicted and simulated values and the fraction of generated structures that meet the target under the independent calculator.
[Table 2; 'Results and Discussion: RL Framework'] The reported statistics do not consistently support the claimed systematic targeting. For CO2 adsorption at 0.1 bar (original mean 0.599, std 0.682), the RL (Mean+2σ) row has mean 0.432 mol/kg, below the original mean and far below the target of 1.963 mol/kg; RL (Mean) is 0.576 vs. target 0.599. For CH4 adsorption at 0.05 bar, RL (Mean) is 0.044 vs. target 0.066, and RL (Mean+2σ) is 0.162 vs. target 0.268. Yet the text states that the framework 'systematically reshape[s] property distributions' and achieves 'targeting performance.' The paper needs a direct success metric—such as the proximity score in Eq. (42), the fraction of generated structures within a stated tolerance of each target, or the per-target percentage that actually reaches the target—before claims of successful targeting can be evaluated.
[Table 1 vs. SI Tables S17-S22; Table 2 label] There is a material inconsistency between the main-text and SI tables. Main-text Table 1 reports for 'CO2 adsorption at 0.01 bar' Mean+2σ validity 100%, novelty 100%, diversity 100%, while SI Table S17 for the same condition reports validity 58.82%, novelty 100%, diversity 99%. The main-text Table 2 lists 'CO2 adsorption at 0.1 bar' as a primary result, although the text identifies CO2 at 0.01 bar as the primary CO2 task; the values in main-text Table 1 match SI Table S22 for 0.1 bar rather than S17 for 0.01 bar. This suggests switched or mislabeled rows and must be resolved, because the 100% validity at the extreme CO2 target is highlighted as a key achievement in the text.
[Stage 2 and Stage 3; 'Fine-tuned Models Fail to Generate Valid MOFs'] The baseline comparison and the RL value network are not clearly defined. Stage 2 extends the base language model with a regression head and describes the result as 'both generator and property evaluator,' while Stage 3 says the value network is the 'frozen property prediction model (MOFormer).' It is therefore unclear whether the 0% validity result attributed to fine-tuned models comes from the same model architecture used for generation, and whether the RL reward uses the fine-tuned GPT's regression head or a separately pretrained MOFormer checkpoint. This matters because the paper's motivation for RL rests on the fine-tuned model's complete failure to generate valid MOFs; the experimental setup and inference procedure for that baseline need to be described precisely.

minor comments (3)

[Table 3] The Target column in Table 3 lists 'High CH4 adsorption at 0.5 bar' and 'High CO2 adsorption at 0.5 bar,' but the primary tasks in the text and Table 1 are CH4 at 0.05/0.9 bar and CO2 at 0.01 bar; please either align the examples with the primary tasks or explicitly state that they come from additional pressure conditions.
[Eq. (6) and SI Table S4] The target reward weight is denoted βtarget in Eq. (6) but αtgt in SI Table S4 and Algorithm 1; please use consistent notation throughout.
[Figure 3 caption] The caption does not state whether the plotted distributions are kernel density estimates of the MOFormer-predicted property values or of ground-truth values; given the closed-loop concern, this should be stated explicitly in the caption and in the text.

Circularity Check

2 steps flagged · score 7.0 of 10

Property-targeting success is measured by the same MOFormer surrogate used as the RL reward, so the reported distribution shifts are the training objective, not independent evidence of improved MOF properties.

self definitional [Stage 3, Eq. (7); Table 2; SI Limitations (S14, S15)]
"Rtarget(m) = ∑ wi·Rproximity(ˆpi(m),Ti) (7) where ˆpi(m) is the predicted property value, Ti is the target value ... The reinforcement learning optimization is only as good as the underlying property prediction model; inaccuracies in property prediction can lead to suboptimal targeting."

The RL policy is updated to maximize Rtarget, which is computed from MOFormer's p_hat values. Table 2 then reports the mean and standard deviation of those same p_hat values as evidence that the framework targets properties. Reporting that the surrogate moved toward the target is restating the reward objective. The only way to break the loop would be independent GCMC, DFT, or experimental evaluation of the generated structures, and none is provided. The SI itself concedes the loop by saying the optimization is only as good as the predictor.
self definitional [Reward Function Design, Eq. (9); Evaluation Metrics, SI Eq. (39)]
"Rvalidity(m) = 1 if valid according to validation procedure, 0 otherwise ... Rvalidity = ∑N i=1 ⊮valid(mi) / N × 100%"

The validity metric is defined by the same validation procedure that is used as the validity reward component during RL training. The reported validity percentages therefore measure whether generated MOFid strings satisfy the reward's own checker, not whether the structures are synthesizable or physically realizable. The abstract's phrase 'synthesizable, topologically valid MOFs' is thus supported by a self-defined check rather than by any 3D structure generation or external validation.

full rationale

The central property-targeting claim is circular in a specific, quotable sense: Eq. (7) defines the target reward from MOFormer's predicted p_hat values, and Table 2 reports the means and standard deviations of those same p_hat values as the outcome statistics. RL maximizes the reward, so a shift in Table 2 is the training objective itself, not an independent confirmation that real MOF adsorption or band gaps improved. The SI Limitations section explicitly concedes that the RL optimization is only as good as the underlying property prediction model. No GCMC, DFT, or experimental validation is reported for any generated structure, so the paper provides no external ground truth to break the loop. The validity claim is similarly self-referential: the validity reward and the validity evaluation use the same validation procedure, so the reported percentages certify compliance with the reward's own checker rather than synthesizability. The method itself, RL over MOFid sequences with multi-component rewards, is not circular and contains independent technical content, and the novelty and diversity metrics are defined outside the reward loop; hence the score is not maximal. However, because the paper's headline claim about 'desired functional attributes' rests entirely on the surrogate that defines success, a score of 7 reflects strong partial circularity.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a fitted surrogate property predictor (used both as RL reward and as evaluation), hand-tuned reward shaping parameters, and string-level validity checks. None of the generated structures is validated by simulation or experiment, and the surrogate's accuracy is not reported. The framework introduces no new physical entities.

free parameters (7)

Reward weights (alpha_tgt, alpha_val, alpha_nov, alpha_div) = 3.0, 2.5, 1.5, 2.0 (SI Table S4)
Hand-chosen to balance target, validity, novelty, and diversity; no sensitivity analysis is reported.
Target reward tier thresholds and values = 15.0/12.0/8.0/4.0 for delta_rel <= 0.05/0.1/0.2/0.5 (SI Eq. 17)
Hand-set reward shaping to distinguish levels of progress toward the target.
Direction bonuses and penalties = 1.3 (achieved), 1.1 (near), 0.95 (wrong direction)
Chosen by experimentation to bias optimization in the desired property direction.
Diversity subweights (wb, wn, wh, wc) = 0.30, 0.25, 0.35, 0.10
Hand-chosen weights for batch, n-gram, history, and composition diversity.
Top-K selection schedule = 0.5 (epochs<100), 0.4 (100-200), 0.3 (>=200)
Adaptive selectivity chosen experimentally to focus learning on promising structures.
Reward normalization thresholds = mean>100 or std>50 triggers normalization; target mean 20, std 10
Hand-set stability parameters from Table S5.
Frozen property predictor (MOFormer) parameters = learned weights from supervised fine-tuning on hMOF/QMOF labels
This fitted surrogate is the only source of property values in the reward and in the evaluation tables; its accuracy is not benchmarked in the paper.

assumptions (4)

domain assumption MOFid string representation captures sufficient chemical and topological information for generating valid MOFs and predicting their properties.
Used throughout; MOFid encodes SBU SMILES and RCSR topology but not 3D coordinates. The SI Limitations states a separate step is needed to generate the full 3D structure, introducing uncertainty.
domain assumption The MOFormer-derived property predictor is accurate enough that optimizing its predicted values yields MOFs with genuinely improved target properties.
This premise carries the entire property-targeting claim. Reward Rtarget (Eq. 7) and evaluation statistics (Table 2) use the same predictor; SI Limitations acknowledges predictor inaccuracies limit targeting quality.
domain assumption The RDKit-with-metal-substitution and RCSR checks constitute a sufficient validity test for generated MOFs.
Validity reward and reported validity rates rely on these string-level checks; no crystallographic reconstruction or synthesis and stability assessment is performed.
domain assumption The aggregate pretraining corpus and the property benchmark splits are representative and do not create leakage between generator pretraining and property evaluation.
Pretraining uses an 80/20 aggregate split over Boyd and Woo, hMOF, and QMOF, while property benchmarks use a separate 0.8/0.05/0.15 split on hMOF and QMOF; the paper does not state that these splits are disjoint, so the generator may have seen property-test structures during pretraining.

how reviews work

0 comments

Cite this review

Pith. "Pith review of MOFGPT: Generative Design of Metal-Organic Frameworks using Language Models." pith.science (2026). https://pith.science/paper/KB6EPFIY

@misc{pith2026250600198,
  author       = {Pith},
  title        = {Pith review of: MOFGPT: Generative Design of Metal-Organic Frameworks using Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KB6EPFIY}},
  note         = {Machine review of arXiv:2506.00198}
}

read the original abstract

The discovery of Metal-Organic Frameworks (MOFs) with application-specific properties remains a central challenge in materials chemistry, owing to the immense size and complexity of their structural design space. Conventional computational screening techniques such as molecular simulations and density functional theory (DFT), while accurate, are computationally prohibitive at scale. Machine learning offers an exciting alternative by leveraging data-driven approaches to accelerate materials discovery. The complexity of MOFs, with their extended periodic structures and diverse topologies, creates both opportunities and challenges for generative modeling approaches. To address these challenges, we present a reinforcement learning-enhanced, transformer-based framework for the de novo design of MOFs. Central to our approach is MOFid, a chemically-informed string representation encoding both connectivity and topology, enabling scalable generative modeling. Our pipeline comprises three components: (1) a generative GPT model trained on MOFid sequences, (2) MOFormer, a transformer-based property predictor, and (3) a reinforcement learning (RL) module that optimizes generated candidates via property-guided reward functions. By integrating property feedback into sequence generation, our method drives the model toward synthesizable, topologically valid MOFs with desired functional attributes. This work demonstrates the potential of large language models, when coupled with reinforcement learning, to accelerate inverse design in reticular chemistry and unlock new frontiers in computational MOF discovery.

Figures

Figures reproduced from arXiv: 2506.00198 by the authors.

**Figure 2.** Reinforcement learning framework for property-driven MOF generation with multi [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

**Figure 3.** Property-targeted generation results across four tasks. (a) CH [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

90 extracted references · 71 canonical work pages

[1]

B.; Bae, Y.-S.; Wilmer, C

Getman, R. B.; Bae, Y.-S.; Wilmer, C. E.; Snurr, R. Q. Review and analysis of molecular simulations of methane, hydrogen, and acetylene storage in metal--organic frameworks. Chemical reviews 2012, 112, 703--723

2012
[2]

G.; Chidambaram, A.; Garc \' a-D \' ez, E.; Ireland, C

Boyd, P. G.; Chidambaram, A.; Garc \' a-D \' ez, E.; Ireland, C. P.; Daff, T. D.; Bounds, R.; G adysiak, A.; Schouwink, P.; Moosavi, S. M.; Maroto-Valer, M. M.; others Data-driven design of metal--organic frameworks for wet flue gas CO2 capture. Nature 2019, 576, 253--256

2019
[3]

Water desalination with two-dimensional metal--organic framework membranes

Cao, Z.; Liu, V.; Barati Farimani, A. Water desalination with two-dimensional metal--organic framework membranes. Nano letters 2019, 19, 8638--8643

2019
[4]

F.; Goscianska, J.; Demessence, A.; Ettlinger, R.; Wuttke, S

Dutta, S.; de Luis, R. F.; Goscianska, J.; Demessence, A.; Ettlinger, R.; Wuttke, S. Metal--organic frameworks for water desalination. Advanced Functional Materials 2024, 34, 2304790

2024
[5]

T.; Peng, L.; Reeder, W

Sun, D. T.; Peng, L.; Reeder, W. S.; Moosavi, S. M.; Tiana, D.; Britt, D. K.; Oveisi, E.; Queen, W. L. Rapid, selective heavy metal removal from water by a metal--organic framework/polydopamine composite. ACS central science 2018, 4, 349--356

2018
[6]

Emerging Multifunctional Metal–Organic Framework Materials

Li, B.; Wen, H.-M.; Cui, Y.; Zhou, W.; Qian, G.; Chen, B. Emerging Multifunctional Metal–Organic Framework Materials. Advanced Materials 2016, 28, 8819--8860

2016
[7]

Methane storage in metal–organic frameworks

He, Y.; Zhou, W.; Qian, G.; Chen, B. Methane storage in metal–organic frameworks. Chem. Soc. Rev. 2014, 43, 5657--5678

2014
[8]

Jiao, L.; Seow, J. Y. R.; Skinner, W. S.; Wang, Z. U.; Jiang, H.-L. Metal–organic frameworks: Structures and functional applications. Materials Today 2019, 27, 43--68

2019

Show all 90 references

[9]

Z.; Li, A.; Wiggin, S

Moghadam, P. Z.; Li, A.; Wiggin, S. B.; Tao, A.; Maloney, A. G. P.; Wood, P. A.; Ward, S. C.; Fairen-Jimenez, D. Development of a Cambridge Structural Database Subset: A Collection of Metal–Organic Frameworks for Past, Present, and Future. Chemistry of Materials 2017, 29, 2618--2625

2017
[10]

E.; Leaf, M.; Lee, C

Wilmer, C. E.; Leaf, M.; Lee, C. Y.; Farha, O. K.; Hauser, B. G.; Hupp, J. T.; Snurr, R. Q. Large-scale screening of hypothetical metal--organic frameworks. Nature chemistry 2012, 4, 83--89

2012
[11]

M.; Ongari, D.; Moosavi, S

Jablonka, K. M.; Ongari, D.; Moosavi, S. M.; Smit, B. Big-Data Science in Porous Materials: Materials Genomics and Machine Learning. Chemical Reviews 2020, 120, 8066--8129, PMID: 32520531

2020
[12]

C.; Slater, B.; Smit, B.; Haranczyk, M

Witman, M.; Ling, S.; Anderson, S.; Tong, L.; Stylianou, K. C.; Slater, B.; Smit, B.; Haranczyk, M. In silico design and screening of hypothetical MOF-74 analogs and their experimental synthesis. Chem. Sci. 2016, 7, 6263--6272

2016
[13]

S.; Notestein, J

Rosen, A. S.; Notestein, J. M.; Snurr, R. Q. Identifying promising metal–organic frameworks for heterogeneous catalysis via high-throughput periodic density functional theory. Journal of Computational Chemistry 2019, 40, 1305--1318

2019
[14]

G.; Chung, Y

Nandy, A.; Yue, S.; Oh, C.; Duan, C.; Terrones, G. G.; Chung, Y. G.; Kulik, H. J. A database of ultrastable MOFs reassembled from stable fragments with machine learning models. Matter 2023, 6, 1585--1603

2023
[15]

D.; Barati Farimani, A

Karamad, M.; Magar, R.; Shi, Y.; Siahrostami, S.; Gates, I. D.; Barati Farimani, A. Orbital graph convolutional neural network for material property prediction. Physical Review Materials 2020, 4, 093801

2020
[16]

M.; Jablonka, K

Moosavi, S. M.; Jablonka, K. M.; Smit, B. The Role of Machine Learning in the Understanding and Design of Materials. Journal of the American Chemical Society 2020, 142, 20273--20287, PMID: 33170678

2020
[17]

B.; Long, J

Islamov, M.; Babaei, H.; Anderson, R.; Sezginel, K. B.; Long, J. R.; McGaughey, A. J.; Gomez-Gualdron, D. A.; Wilmer, C. E. High-throughput screening of hypothetical metal-organic frameworks for thermal conductivity. npj Computational Materials 2023, 9, 11

2023
[18]

S.; Notestein, J

Rosen, A. S.; Notestein, J. M.; Snurr, R. Q. Realizing the data-driven, computational discovery of metal-organic framework catalysts. Current Opinion in Chemical Engineering 2022, 35, 100760

2022
[19]

From Data to Discovery: Recent Trends of Machine Learning in Metal–Organic Frameworks

Park, J.; Kim, H.; Kang, Y.; Lim, Y.; Kim, J. From Data to Discovery: Recent Trends of Machine Learning in Metal–Organic Frameworks. JACS Au 2024, 4, 3727--3743

2024
[20]

MOFormer: Self-Supervised Transformer Model for Metal–Organic Framework Property Prediction

Cao, Z.; Magar, R.; Wang, Y.; Barati Farimani, A. MOFormer: Self-Supervised Transformer Model for Metal–Organic Framework Property Prediction. Journal of the American Chemical Society 2023, 145, 2958--2967, PMID: 36706365

2023
[21]

Harnessing Large Language Models to Collect and Analyze Metal--Organic Framework Property Data Set

Kang, Y.; Lee, W.; Bae, T.; Han, S.; Jang, H.; Kim, J. Harnessing Large Language Models to Collect and Analyze Metal--Organic Framework Property Data Set. Journal of the American Chemical Society 2025,

2025
[22]

Evaluation of open-source large language models for metal--organic frameworks research

Bai, X.; Xie, Y.; Zhang, X.; Han, H.; Li, J.-R. Evaluation of open-source large language models for metal--organic frameworks research. Journal of Chemical Information and Modeling 2024, 64, 4958--4965

2024
[23]

Radford, A.; Narasimhan, K.; Salimans, T.; Sutskever, I.; others Improving language understanding by generative pre-training. 2018,

2018
[24]

Brown, T. B. et al. Language Models are Few-Shot Learners. 2020,

2020
[25]

Crystal twins: self-supervised learning for crystalline material property prediction

Magar, R.; Wang, Y.; Barati Farimani, A. Crystal twins: self-supervised learning for crystalline material property prediction. npj Computational Materials 2022, 8, 231

2022
[26]

C.; Aksu, G

Demir, H.; Daglar, H.; Gulbalkan, H. C.; Aksu, G. O.; Keskin, S. Recent advances in computational modeling of MOFs: From molecular simulations to machine learning. Coordination Chemistry Reviews 2023, 484, 215112

2023
[27]

F.; Keskin, S.; Yildirim, R

Altintas, C.; Altundal, O. F.; Keskin, S.; Yildirim, R. Machine Learning Meets with Metal Organic Frameworks for Gas Storage and Separation. Journal of Chemical Information and Modeling 2021, 61, 2131--2146, PMID: 33914526

2021
[28]

W.; Kulik, H

Nandy, A.; Terrones, G.; Arunachalam, N.; Duan, C.; Kastner, D. W.; Kulik, H. J. MOFSimplify, machine learning models with extracted stability data of three thousand metal--organic frameworks. Scientific Data 2022, 9, 74

2022
[29]

Accelerating CO₂ Direct Air Capture Screening for Metal-Organic Frameworks with a Transferable Machine Learning Force Field

Lim, Y.; Park, H.; Walsh, A.; Kim, J. Accelerating CO₂ Direct Air Capture Screening for Metal-Organic Frameworks with a Transferable Machine Learning Force Field. 2024,

2024
[30]

Crystal Diffusion Variational Autoencoder for Periodic Material Generation

Xie, T.; Fu, X.; Ganea, O.-E.; Barzilay, R.; Jaakkola, T. Crystal Diffusion Variational Autoencoder for Periodic Material Generation. arXiv preprint arXiv:2110.06197 2021,

2021 arXiv
[31]

Application of Transformers in Cheminformatics

Luong, K.-D.; Singh, A. Application of Transformers in Cheminformatics. Journal of Chemical Information and Modeling 2024, 64, 4392--4409, PMID: 38815246

2024
[32]

Multimodal language and graph learning of adsorption configuration in catalysis

Ock, J.; Badrinarayanan, S.; Magar, R.; Antony, A.; Barati Farimani, A. Multimodal language and graph learning of adsorption configuration in catalysis. Nature Machine Intelligence 2024, 1--11

2024
[33]

PeptideBERT: A Language Model Based on Transformers for Peptide Property Prediction

Guntuboina, C.; Das, A.; Mollaei, P.; Kim, S.; Barati Farimani, A. PeptideBERT: A Language Model Based on Transformers for Peptide Property Prediction. The Journal of Physical Chemistry Letters 2023, 14, 10427--10434, PMID: 37956397

2023
[34]

Catalyst Energy Prediction with CatBERTa: Unveiling Feature Exploration Strategies through Large Language Models

Ock, J.; Guntuboina, C.; Barati Farimani, A. Catalyst Energy Prediction with CatBERTa: Unveiling Feature Exploration Strategies through Large Language Models. ACS Catalysis 2023, 13, 16032--16044

2023
[35]

H.; Rieth, A

Hendon, C. H.; Rieth, A. J.; Korzyński, M. D.; Dincă, M. Grand Challenges and Future Opportunities for Metal–Organic Frameworks. ACS Central Science 2017, 3, 554--563, PMID: 28691066

2017
[36]

T.; Huisman, B

Gupta, A.; Müller, A. T.; Huisman, B. J. H.; Fuchs, J. A.; Schneider, P.; Schneider, G. Generative Recurrent Networks for De Novo Drug Design. Molecular Informatics 2018, 37, 1700111

2018
[37]

The problem of learning long-term dependencies in recurrent networks

Bengio, Y.; Frasconi, P.; Simard, P. The problem of learning long-term dependencies in recurrent networks. IEEE International Conference on Neural Networks. 1993; pp 1183--1188 vol.3

1993
[38]

De novo generation of hit-like molecules from gene expression signatures using artificial intelligence

M \'e ndez-Lucio, O.; Baillif, B.; Clevert, D.-A.; Rouqui \'e , D.; Wichard, J. De novo generation of hit-like molecules from gene expression signatures using artificial intelligence. Nature communications 2020, 11, 10

2020
[39]

Conditional Molecular Design with Deep Generative Models

Kang, S.; Cho, K. Conditional Molecular Design with Deep Generative Models. Journal of Chemical Information and Modeling 2019, 59, 43--52, PMID: 30016587

2019
[40]

Communications Materials 2022, 3, 93

Reiser, P.; Neubert, M.; Eberhard, A.; Torresi, L.; Zhou, C.; Shao, C.; Metni, H.; van Hoesel, C.; Schopmans, H.; Sommer, T.; others Graph neural networks for materials science and chemistry. Communications Materials 2022, 3, 93

2022
[41]

A.; Chaudhuri, S.; Cooper, D.; Foster, I.; Tajkhorshid, E

Park, H.; Yan, X.; Zhu, R.; Huerta, E. A.; Chaudhuri, S.; Cooper, D.; Foster, I.; Tajkhorshid, E. A generative artificial intelligence framework based on a molecular diffusion model for the design of metal-organic frameworks for carbon capture. Communications Chemistry 2024, 7, 21

2024
[42]

J.; Yang, S.; Kaplan, A.; hsu Lin, Y.; Yin, J.; Mirzaei, S.; Abdelgaid, M.; Alawadhi, A

Inizan, T. J.; Yang, S.; Kaplan, A.; hsu Lin, Y.; Yin, J.; Mirzaei, S.; Abdelgaid, M.; Alawadhi, A. H.; Cho, K.; Zheng, Z.; Cubuk, E. D.; Borgs, C.; Chayes, J. T.; Persson, K. A.; Yaghi, O. M. System of Agentic AI for the Discovery of Metal-Organic Frameworks. 2025; https://ar...

2025 arXiv
[43]

S.; Jaakkola, T

Fu, X.; Xie, T.; Rosen, A. S.; Jaakkola, T. S.; Smith, J. A. MOFD iff: Coarse-grained Diffusion for Metal-Organic Framework Design. The Twelfth International Conference on Learning Representations. 2024

2024
[44]

Park, J.; Gill, A. P. S.; Moosavi, S. M.; Kim, J. Inverse design of porous materials: a diffusion model approach. J. Mater. Chem. A 2024, 12, 6507--6514

2024
[45]

Kotsias, P.-C.; Ar \'u s-Pous, J.; Chen, H.; Engkvist, O.; Tyrchan, C.; Bjerrum, E. J. Direct steering of de novo molecular generation with descriptor conditional recurrent neural networks. Nature Machine Intelligence 2020, 2, 254--265

2020
[46]

Bidirectional Molecule Generation with Recurrent Neural Networks

Grisoni, F.; Moret, M.; Lingwood, R.; Schneider, G. Bidirectional Molecule Generation with Recurrent Neural Networks. Journal of Chemical Information and Modeling 2020, 60, 1175--1183, PMID: 31904964

2020
[47]

S.; Bucior, B

Yao, Z.; S \'a nchez-Lengeling, B.; Bobbitt, N. S.; Bucior, B. J.; Kumar, S. G. H.; Collins, S. P.; Burns, T.; Woo, T. K.; Farha, O. K.; Snurr, R. Q.; others Inverse design of nanoporous crystalline reticular materials with deep generative models. Nature Machine Intelligence 2...

2021
[48]

P.; Bengio, Y.; Li, W

Che, T.; Li, Y.; Jacob, A. P.; Bengio, Y.; Li, W. Mode Regularized Generative Adversarial Networks. ArXiv 2016, abs/1612.02136

2016 arXiv
[49]

Are VAEs Bad at Reconstructing Molecular Graphs? 2023; https://arxiv.org/abs/2305.03041

Muenkler, H.; Misztela, H.; Pikusa, M.; Segler, M.; Schneider, N.; Maziarz, K. Are VAEs Bad at Reconstructing Molecular Graphs? 2023; https://arxiv.org/abs/2305.03041

2023 arXiv
[50]

M.; Nandy, A.; Jablonka, K

Moosavi, S. M.; Nandy, A.; Jablonka, K. M.; Ongari, D.; Janet, J. P.; Boyd, P. G.; Lee, Y.; Smit, B.; Kulik, H. J. Understanding the diversity of the metal-organic framework ecosystem. Nature communications 2020, 11, 4068

2020
[51]

Has generative artificial intelligence solved inverse materials design? Matter 2024, 7, 2355--2367

Park, H.; Li, Z.; Walsh, A. Has generative artificial intelligence solved inverse materials design? Matter 2024, 7, 2355--2367

2024
[52]

N.; Kaiser, .; Polosukhin, I

Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A. N.; Kaiser, .; Polosukhin, I. Attention is all you need. Advances in neural information processing systems 2017, 30

2017
[53]

Language Models are Unsupervised Multitask Learners

Radford, A.; Wu, J.; Child, R.; Luan, D.; Amodei, D.; Sutskever, I. Language Models are Unsupervised Multitask Learners. 2019,

2019
[54]

K.; Priyakumar, U

Bagal, V.; Aggarwal, R.; Vinod, P. K.; Priyakumar, U. D. MolGPT: Molecular Generation Using a Transformer-Decoder Model. Journal of Chemical Information and Modeling 2022, 62, 2064--2076, PMID: 34694798

2022
[55]

J.; Rosen, A

Bucior, B. J.; Rosen, A. S.; Haranczyk, M.; Yao, Z.; Ziebel, M. E.; Farha, O. K.; Hupp, J. T.; Siepmann, J. I.; Aspuru-Guzik, A.; Snurr, R. Q. Identification Schemes for Metal–Organic Frameworks To Enable Rapid Search and Cheminformatics Analysis. Crystal Growth & Design 2019,...

2019
[56]

SMILES, a chemical language and information system

Weininger, D. SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules. Journal of Chemical Information and Computer Sciences 1988, 28, 31--36

1988
[57]

A.; Ramsden, S

O’Keeffe, M.; Peskov, M. A.; Ramsden, S. J.; Yaghi, O. M. The Reticular Chemistry Structure Resource (RCSR) Database of, and Symbols for, Crystal Nets. Accounts of Chemical Research 2008, 41, 1782--1789, PMID: 18834152

2008
[58]

Inverse design of porous materials using artificial neural networks

Kim, B.; Lee, S.; Kim, J. Inverse design of porous materials using artificial neural networks. Science Advances 2020, 6, eaax9324

2020
[59]

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

DeepSeek-AI et al. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. 2025; https://arxiv.org/abs/2501.12948

2025 arXiv
[60]

Reinforced Adversarial Neural Computer for de Novo Molecular Design

Putin, E.; Asadulaev, A.; Ivanenkov, Y.; Aladinskiy, V.; Sanchez-Lengeling, B.; Aspuru-Guzik, A.; Zhavoronkov, A. Reinforced Adversarial Neural Computer for de Novo Molecular Design. Journal of Chemical Information and Modeling 2018, 58, 1194--1204, PMID: 29762023

2018
[61]

Deep reinforcement learning for de novo drug design

Popova, M.; Isayev, O.; Tropsha, A. Deep reinforcement learning for de novo drug design. Science Advances 2018, 4, eaap7885

2018
[62]

Molecule generation using transformers and policy gradient reinforcement learning

Mazuz, E.; Shtar, G.; Shapira, B.; Rokach, L. Molecule generation using transformers and policy gradient reinforcement learning. Scientific Reports 2023, 13, 8799

2023
[63]

Deep reinforcement learning for inverse inorganic materials design

Pan, E.; Karpovich, C.; Olivetti, E. Deep reinforcement learning for inverse inorganic materials design. arXiv preprint arXiv:2210.11931 2022,

2022 arXiv
[64]

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

Park, H.; Majumdar, S.; Zhang, X.; Kim, J.; Smit, B. Inverse design of metal--organic frameworks for direct air capture of CO 2 via deep reinforcement learning. Digital Discovery 2024, 3, 728--741

2024
[65]

T.; Yaghi, O

Zheng, Z.; Rong, Z.; Rampal, N.; Borgs, C.; Chayes, J. T.; Yaghi, O. M. A GPT-4 Reticular Chemist for Guiding MOF Discovery. Angewandte Chemie International Edition 2023, 62, e202311983

2023
[66]

ChatMOF: an artificial intelligence system for predicting and generating metal-organic frameworks using large language models

Kang, Y.; Kim, J. ChatMOF: an artificial intelligence system for predicting and generating metal-organic frameworks using large language models. Nature Communications 2024, 15, 4705

2024
[67]

S.; Iyer, S

Rosen, A. S.; Iyer, S. M.; Ray, D.; Yao, Z.; Aspuru-Guzik, A.; Gagliardi, L.; Notestein, J. M.; Snurr, R. Q. Machine learning the quantum-chemical properties of metal–organic frameworks for accelerated materials discovery. Matter 2021, 4, 1578--1597

2021
[68]

S.; Fung, V.; Huck, P.; O’Donnell, C

Rosen, A. S.; Fung, V.; Huck, P.; O’Donnell, C. T.; Horton, M. K.; Truhlar, D. G.; Persson, K. A.; Notestein, J. M.; Snurr, R. Q. High-throughput predictions of metal--organic framework electronic properties: theoretical challenges, graph neural networks, and data exploration....

2022
[69]

S.; McAllester, D.; Singh, S.; Mansour, Y

Sutton, R. S.; McAllester, D.; Singh, S.; Mansour, Y. Policy Gradient Methods for Reinforcement Learning with Function Approximation. Advances in Neural Information Processing Systems. 1999

1999
[70]

G.; Veenstra, M.; Matzger, A

Ahmed, A.; Seth, S.; Purewal, J.; Wong-Foy, A. G.; Veenstra, M.; Matzger, A. J.; Siegel, D. J. Exceptional hydrogen storage achieved by screening nearly half a million metal-organic frameworks. Nature communications 2019, 10, 1568

2019
[71]

G.; Camp, J.; Haranczyk, M.; Sikora, B

Chung, Y. G.; Camp, J.; Haranczyk, M.; Sikora, B. J.; Bury, W.; Krungleviciute, V.; Yildirim, T.; Farha, O. K.; Sholl, D. S.; Snurr, R. Q. Computation-ready, experimental metal--organic frameworks: A tool to enable high-throughput screening of nanoporous crystals. Chemistry of...

2014
[72]

A.; Sumida, K.; Herm, Z

Mason, J. A.; Sumida, K.; Herm, Z. R.; Krishna, R.; Long, J. R. Evaluating metal--organic frameworks for post-combustion carbon dioxide capture via temperature swing adsorption. Energy & Environmental Science 2011, 4, 3030--3040

2011
[73]

H.; Park, S

Sun, L.; Hendon, C. H.; Park, S. S.; Tulchinsky, Y.; Wan, R.; Wang, F.; Walsh, A.; Dinc a , M. Is iron unique in promoting electrical conductivity in MOFs? Chemical science 2017, 8, 4450--4457

2017
[74]

Q.; Mai, T.; Pham-Tran, N.-N.; Kawazoe, Y.; Mizuseki, H.; Nguyen-Manh, D

Pham, H. Q.; Mai, T.; Pham-Tran, N.-N.; Kawazoe, Y.; Mizuseki, H.; Nguyen-Manh, D. Engineering of band gap in metal--organic frameworks by functionalizing organic linker: A systematic density functional theory investigation. The Journal of Physical Chemistry C 2014, 118, 4567--4577

2014
[75]

He, Yabing and Zhou, Wei and Qian, Guodong and Chen, Banglin

Narayan, T. C.; Miyakai, T.; Seki, S.; Dincă, M. High charge mobility in a tetrathiafulvalene-based microporous metal--organic framework. Journal of the American Chemical Society 2012, 134, 12932--12935 mcitethebibliography main.bib00006640000000000000000000152435150164072670...

2012
[76]

A.; Bekas, C.; Lee, A

Schwaller, P.; Laino, T.; Gaudin, T.; Bolgar, P.; Hunter, C. A.; Bekas, C.; Lee, A. A. Molecular Transformer: A Model for Uncertainty-Calibrated Chemical Reaction Prediction. ACS Central Science 2019, 5, 1572--1583, PMID: 31572784

2019
[77]

Found in Translation

Schwaller, P.; Gaudin, T.; Lányi, D.; Bekas, C.; Laino, T. “Found in Translation”: predicting outcomes of complex organic chemistry reactions using neural sequence-to-sequence models. Chem. Sci. 2018, 9, 6091--6098

2018
[78]

C.; Nair, V

Schwaller, P.; Probst, D.; Vaucher, A. C.; Nair, V. H.; Kreutter, D.; Laino, T.; Reymond, J.-L. Mapping the space of chemical reactions using attention-based neural networks. Nature Machine Intelligence 2021, 3, 144--152

2021
[80]

S.; Skorupskii, G.; Dincă, M

Xie, L. S.; Skorupskii, G.; Dincă, M. Electrically Conductive Metal–Organic Frameworks. Chemical Reviews 2020, 120, 8536--8580, PMID: 32275412

2020
[81]

C.; Elias, J

Sheberla, D.; Bachman, J. C.; Elias, J. S.; Sun, C.-J.; Shao-Horn, Y.; Dinc a , M. Conductive MOF electrodes for stable supercapacitors with high areal capacitance. Nature materials 2017, 16, 220--224

2017
[82]

Identification of optimal metal-organic frameworks by machine learning: Structure decomposition, feature integration, and predictive modeling

Wang, Z.; Zhou, Y.; Zhou, T.; Sundmacher, K. Identification of optimal metal-organic frameworks by machine learning: Structure decomposition, feature integration, and predictive modeling. Computers & Chemical Engineering 2022, 160, 107739

2022
[83]

V.; Soldatov, M

Butova, V. V.; Soldatov, M. A.; Guda, A. A.; Lomachenko, K. A.; Lamberti, C. Metal-organic frameworks: structure, properties, methods of synthesis and characterization. Russian Chemical Reviews 2016, 85, 280

2016
[84]

Proximal Policy Optimization Algorithms

Schulman, J.; Wolski, F.; Dhariwal, P.; Radford, A.; Klimov, O. Proximal Policy Optimization Algorithms. arXiv preprint arXiv:1707.06347 2017,

2017 arXiv
[85]

Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor

Haarnoja, T.; Zhou, A.; Abbeel, P.; Levine, S. Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor. arXiv preprint arXiv:1801.01290 2018,

2018 arXiv
[86]

M.; Jablonka, K

Majumdar, S.; Moosavi, S. M.; Jablonka, K. M.; Ongari, D.; Smit, B. Diversifying Databases of Metal Organic Frameworks for High-Throughput Computational Screening. ACS Applied Materials & Interfaces 2021, 13, 61004--61014, PMID: 34910455

2021
[87]

Metal–organic frameworks for artificial photosynthesis and photocatalysis

Zhang, T.; Lin, W. Metal–organic frameworks for artificial photosynthesis and photocatalysis. Chem. Soc. Rev. 2014, 43, 5982--5993

2014
[88]

E.; O’Keeffe, M.; Yaghi, O

Furukawa, H.; Cordova, K. E.; O’Keeffe, M.; Yaghi, O. M. The Chemistry and Applications of Metal-Organic Frameworks. Science 2013, 341, 1230444

2013
[89]

Metal-Organic Framework Composites for Catalysis

Chen, L.; Xu, Q. Metal-Organic Framework Composites for Catalysis. Matter 2019, 1, 57--89

2019
[90]

D.; Stavila, V

Allendorf, M. D.; Stavila, V. Crystal engineering, structure-function relationships, and the future of metal-organic frameworks. CrystEngComm 2015, 17, 229 – 246, Cited by: 233; All Open Access, Green Open Access

2015
[91]

Schwaller, Philippe and Gaudin, Théophile and Lányi, Dávid and Bekas, Costas and Laino, Teodoro

Safaei, M.; Foroughi, M. M.; Ebrahimpoor, N.; Jahani, S.; Omidi, A.; Khatami, M. A review on metal-organic frameworks: Synthesis and applications. TrAC Trends in Analytical Chemistry 2019, 118, 401--425 mcitethebibliography si.bib00006640000000000000000000077441150164072670106...

2019 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

B.; Bae, Y.-S.; Wilmer, C

Getman, R. B.; Bae, Y.-S.; Wilmer, C. E.; Snurr, R. Q. Review and analysis of molecular simulations of methane, hydrogen, and acetylene storage in metal--organic frameworks. Chemical reviews 2012, 112, 703--723

2012

[2] [2]

G.; Chidambaram, A.; Garc \' a-D \' ez, E.; Ireland, C

Boyd, P. G.; Chidambaram, A.; Garc \' a-D \' ez, E.; Ireland, C. P.; Daff, T. D.; Bounds, R.; G adysiak, A.; Schouwink, P.; Moosavi, S. M.; Maroto-Valer, M. M.; others Data-driven design of metal--organic frameworks for wet flue gas CO2 capture. Nature 2019, 576, 253--256

2019

[3] [3]

Water desalination with two-dimensional metal--organic framework membranes

Cao, Z.; Liu, V.; Barati Farimani, A. Water desalination with two-dimensional metal--organic framework membranes. Nano letters 2019, 19, 8638--8643

2019

[4] [4]

F.; Goscianska, J.; Demessence, A.; Ettlinger, R.; Wuttke, S

Dutta, S.; de Luis, R. F.; Goscianska, J.; Demessence, A.; Ettlinger, R.; Wuttke, S. Metal--organic frameworks for water desalination. Advanced Functional Materials 2024, 34, 2304790

2024

[5] [5]

T.; Peng, L.; Reeder, W

Sun, D. T.; Peng, L.; Reeder, W. S.; Moosavi, S. M.; Tiana, D.; Britt, D. K.; Oveisi, E.; Queen, W. L. Rapid, selective heavy metal removal from water by a metal--organic framework/polydopamine composite. ACS central science 2018, 4, 349--356

2018

[6] [6]

Emerging Multifunctional Metal–Organic Framework Materials

Li, B.; Wen, H.-M.; Cui, Y.; Zhou, W.; Qian, G.; Chen, B. Emerging Multifunctional Metal–Organic Framework Materials. Advanced Materials 2016, 28, 8819--8860

2016

[7] [7]

Methane storage in metal–organic frameworks

He, Y.; Zhou, W.; Qian, G.; Chen, B. Methane storage in metal–organic frameworks. Chem. Soc. Rev. 2014, 43, 5657--5678

2014

[8] [8]

Jiao, L.; Seow, J. Y. R.; Skinner, W. S.; Wang, Z. U.; Jiang, H.-L. Metal–organic frameworks: Structures and functional applications. Materials Today 2019, 27, 43--68

2019