Pith. sign in

REVIEW 4 major objections 5 minor 27 references

Improved Molecular Generation through Attribute-Driven Integrative Embeddings and GAN Selectivity

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Adding a classifier-based range-loss to a molecular GAN's generator makes it output odorant molecules 99.2% of the time, versus 52% for a vanilla GAN trained only on odorants, without sacrificing novelty or uniqueness.

desk verdict A coherent pipeline with a credible 94% reconstruction result, but the headline selectivity number is measured by the same classifier that steers the generator, so the central claim is not yet supported. read the letter →

arxiv 2504.19040 v1 pith:EXRXUYE4 submitted 2025-04-26 cs.LG cs.AI

Nandan Joshi , Erhan Guven This is my paper · ORCID
classification cs.LGcs.AI
keywords moleculargenerationgenerativeadversarialnetworkstransformerembeddingsdescriptorsrange-lossodorantmoleculesSMILESreconstructionselective
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

The paper argues that you can steer a molecular GAN toward a target property by adding a simple penalty to the generator's loss, provided the molecules are represented as dense embeddings that a transformer can map back to valid SMILES strings. Its test case is odorancy: the modified range-loss GAN produces molecules that its own odorant classifier labels as odorants 99.2% of the time, versus 52% for a vanilla GAN trained only on odorant examples. The embedding pathway is a transformer encoder that turns circular fingerprints plus 30 global molecular descriptors into 150x512 vectors, with a decoder that reconstructs the original SMILES with 94% accuracy. If the result holds, property-conditioned molecular generation becomes a small modification to an existing GAN rather than a redesigned architecture.

What carries the argument

The central object is the range-loss function $L_{\mathrm{rng}}$ of Eq. (2)-(3), added to the WGAN generator objective as $\min_G (-\mathbb{E}_{z\sim P_z}[D(G(z))] + \lambda_1 L_{\mathrm{rng}})$. The range-loss is defined so that only samples outside the desired property interval contribute a penalty, with a smooth negative-log-likelihood-style term whose gradient vanishes inside the interval and grows for far-outside samples; since odorancy is binary, the paper computes the penalty from the binary cross-entropy of the odorant classifier, treating the target interval as $[0.5,1]$. The other load-bearing piece is the 150x512 vector embedding: a transformer encoder maps the descriptor (circular fingerprints plus 30 global molecular attributes) to this embedding, and a transformer decoder reconstructs SMILES from it, so any new embedding produced by the generator can be translated into a molecule.

What would settle it

Take the molecules the range-loss GAN labels as odorants and score them with a completely independent odorant predictor, or submit them to experimental olfactory testing; if substantially fewer than 99% are odorants by that measure, the reported selectivity is an artifact of the classifier rather than chemistry. A cheaper check: retrain the same pipeline with the classifier's labels randomly shuffled, and see whether the generator still achieves near-perfect selectivity by the shuffled oracle.

Watch

Extended reading notes

Core claim

The central claim is that combining a discriminative classifier's output with the generator's Wasserstein loss, through a range-loss that penalizes only samples outside the target property interval, forces a molecular GAN to generate almost exclusively molecules with the desired property. In the odorant test case, 99.2% of the generated molecules are classified as odorants, compared with 52% for a vanilla GAN trained only on odorant molecules, while uniqueness (99%) and novelty (100%) are essentially unchanged and validity remains at 62%. The paper attributes this selectivity to the range-loss gradient flowing from the odorant classifier through the generator's embedding output, and to the fact that those embeddings are decoded into valid SMILES strings by a pre-trained transformer decoder.

Load-bearing premise

The odorant classifier used both to steer the generator and to judge its outputs is a trustworthy measure of real odorancy; the paper validates it only against the same label set used to train it.

Editorial extensions

If this is right

  • Adding range-loss to a GAN generator is enough to swing output selectivity from 52% to 99.2% on the odorant task, without a dedicated conditional architecture.
  • The transformer encoder-decoder can reconstruct SMILES from the integrative descriptor at 94% accuracy, so generated embeddings are not dead vectors; they correspond to readable molecules.
  • Selectivity comes at little cost to standard GAN metrics: 99% uniqueness, 100% novelty, and 62% validity in the range-loss run, comparable to the vanilla GAN.
  • Because the framework conditions on a single scalar property, extending it to multiple attributes would require adding one range-loss term per property or a vector-valued classifier.

Reading between the lines

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

  • Because the same classifier both supplies the training gradient and scores the outputs, the reported 99.2% likely measures how well the generator satisfies the classifier, not necessarily how odors a human would smell; an independent olfactory or orthogonal model validation would be needed to separate the two.
  • A range-loss built on a differentiable classifier is susceptible to shortcut learning: the generator may discover embeddings that fool the classifier without corresponding to real chemical odorancy. A testable extension is to swap the classifier's penultimate layer or add a second independently trained classifier and check whether selectivity persists.
  • The same descriptor-plus-transformer pipeline could be applied to other property labels such as bioactivity, solubility, or toxicity, with range-loss replacing the domain-specific classifier; if the mechanism is general, the 99%-level selectivity should transfer.
  • The 52% result for the vanilla GAN trained only on odorants hints that even property-enriched training sets do not guarantee selective generation, which is why the explicit loss modification matters.
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

4 major / 5 minor

Summary. The paper proposes a pipeline for property-conditioned molecular generation. A transformer encoder-decoder is trained to map a descriptor (Morgan fingerprint plus 30 RDKit features) to a 150x512 embedding and back to SMILES, achieving 94% reconstruction accuracy. A Wasserstein GAN is then trained on these real embeddings, and its generator loss is augmented with a range-loss term computed from the binary cross-entropy of a separately trained odorant/non-odorant classifier (O-1). The authors report that the range-loss GAN produces 99.2% odorant molecules per O-1, versus 52% for a vanilla GAN trained on an O-1-filtered odorant-only subset of ZINC, and that novelty and uniqueness are 100% with 62% SMILES validity. The central claim, stated in the abstract, is that the modified GAN 'exclusively generates odorant molecules.'

Significance. If the results were supported, the paper would offer a simple and useful recipe for steering molecular GANs toward a desired property: a classifier-based range-loss added to the WGAN objective, combined with a transformer embedding that is decodable to SMILES. The proposed descriptor and the use of a frozen decoder for translating generated embeddings are sensible and potentially valuable. However, the key evidence for the headline claim is currently circular: the same classifier provides the training signal, filters the baseline training set, and scores the generated molecules. No independent validation of odorancy is provided, and the reported numbers themselves contradict the 'exclusively generates' wording. The paper's methodological contribution is plausible, but the experimental validation is not yet sufficient to establish it.

major comments (4)
  1. [Section II.C and Table II] The evaluation of selectivity is circular. The range-loss in Eq. (1)-(3) is computed from the binary cross-entropy of the odorant classifier O-1, and Table II reports the proportion of generated molecules classified as odorants by this same O-1. Additionally, the vanilla GAN baseline is trained on a ZINC subset that was itself filtered by O-1 (Section III.D). The comparison therefore measures how well the generator optimizes O-1's score, not how odorant the generated molecules are in any independent sense. To support the claim, the authors need an external validation: for example, an independent odorant classifier trained on a disjoint dataset, or experimental or literature-based confirmation for a sample of generated molecules, and a demonstration that generated embeddings lie on the real embedding manifold (e.g., distance to nearest real embedding).
  2. [Abstract and Table II] The abstract's statement that 'the GAN exclusively generates odorant molecules' is not supported by the paper's own numbers. Table II reports 99.2%, not 100%, and Table III reports 62% SMILES validity, meaning 38% of generated outputs are not valid molecules and thus cannot be odorants. The claim should be qualified to, for example, 'almost all valid generated molecules are classified as odorants by the in-house classifier.'
  3. [Section IV.C and Table III] All reported GAN metrics (99.2%, 52%, 99%, 62%, 100%, 0.81, 0.77) are single-run point estimates with no indication of the number of generated samples, no multiple seeds, and no variance or confidence intervals. Without this information, the 99.2% versus 52% difference cannot be assessed for statistical significance, and the claim that range-loss does not degrade uniqueness or novelty is not established.
  4. [Section II.C.1] The adaptation of the range-loss from regression to binary classification is under-specified. Eq. (2)-(3) define a continuous range-loss, but the text says only that 'the binary cross-entropy of the odorant classifier output is considered for the range-loss calculation, with its intended range lying between 0.5 and 1.' The exact form of L_rng in terms of the classifier's BCE is not given, so the reader cannot reproduce the training objective.
minor comments (5)
  1. [Section I.A] The SMILES reference [9] has an incomplete author name ('eininger, David' should be 'Weininger, David').
  2. [Reference [10]] The DOI for the transformer paper is incorrect; it appears to be a copy of the SMILES reference's DOI. Please correct it.
  3. [Section III.D] The text reads 'labeled data from the DeepOlf model,labeled data from the DeepOlf model' with a duplicated phrase; this should be cleaned up.
  4. [Section III.D] The abbreviation 'OD/NOD' is used without definition; it presumably means odorant/non-odorant and should be spelled out at first use.
  5. [Section III.A] The Adam optimizer hyperparameters are given as 'beta1 and beta2 of 0.1 and 0.99'; this is unusual (typical beta1 is 0.9) and should be double-checked.

Circularity Check

2 steps flagged · score 7.0 of 10

The 99.2% odorant selectivity is measured by the same O-1 classifier whose BCE is the range-loss training signal, so the headline result reduces to optimizing O-1 rather than demonstrating real odorant generation.

  1. self definitional [Section II.C.1 (Eqs. 1-3) and Section IV.C (Table II); conclusion in Section V]
    "The newly generated vector embeddings are passed to an OD/NOD classifier (O-1), and the range-loss reflects the proportion of non-odorant samples. By minimizing this loss, the generator is thus effectively trained to generate only odorants. ... the binary cross-entropy (BCE) of the odorant classifier output is considered for the range-loss calculation, with its intended range lying between 0.5 and 1."

    The selective-generation objective is literally O-1's BCE: L_rng in Eq. (1)-(3) is computed from O-1's probability, targeted to [0.5,1]. Table II's '% Odorants' is the fraction of generated molecules that O-1 labels odorant, as the conclusion admits: 'generates almost all odorants based on their classification.' Thus the GAN is trained to maximize O-1's positive score and then scored by that same score; the 99.2% result is a property of the optimization loop, not an independent demonstration of odorancy. O-1 was trained on external DeepOlf labels and achieves 95% holdout accuracy, so O-1 has some independent grounding; but no generated molecule is tested against DeepOlf or any other external olfactory benchmark, so the headline selectivity claim is not externally validated.

  2. self definitional [Section III.D (datasets) and Section IV.C (Table II)]
    "To obtain the odorant-only dataset, the ZINC 2.6M dataset is filtered using the OD/NOD classifier to identify only the odorants, and a reduced dataset of 13555 molecules is obtained."

    The vanilla-GAN comparator is trained on a ZINC subset that O-1 selected as odorant, and its output is then scored in Table II by the proportion of O-1 positives. Because both the training distribution and the evaluation metric come from O-1, the 52% baseline and the 99.2% range-loss number are measured entirely inside O-1's labeling; the comparison shows that optimizing O-1's BCE raises O-1's positive rate, not that either model generates compounds with independently verified odorancy.

full rationale

The transformer reconversion (94% on a ZINC holdout) and the odorant classifier's own metrics (95% accuracy on a DeepOlf holdout) are externally anchored and not circular. The circularity is confined to the paper's headline selectivity claim. Eq. (1)-(3) define the generator's extra loss from O-1's BCE with target range [0.5,1]; Section II.C states that the range-loss reflects the proportion of non-odorant samples. Table II then reports the 99.2% 'Proportion of odorants produced' using that same classifier's labels, as the conclusion states ('based on their classification'). The vanilla-GAN baseline is also filtered and measured by O-1. Consequently, the headline result is a closed loop: the objective is O-1's probability, and the measure is O-1's probability. The result would be informative only if O-1 were independently validated on generated molecules, e.g., against DeepOlf labels or expert olfactory annotation, which the paper does not provide. This is a structural circularity of the central claim, not a matter of author intent or citation practice; no load-bearing self-citation is involved. Score 7 reflects that O-1 has some external training signal, so the loop is not a pure tautology, but the reported evidence does not break it.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on two measurement loops: the transformer reconstruction shows the descriptor is expressive, and the GAN selectivity shows the range-loss works. The first loop is empirically supported at 94%, though no error bars are given. The second loop is closed through classifier O-1, which acts as training reward, dataset filter, and evaluation metric. Most free parameters are hand-set hyperparameters of the range-loss; no new physical entities are introduced.

free parameters (5)
  • range_loss_weight_lambda_1 = 10
    Coefficient on L_rng in Eq. 1; chosen by hand to balance WGAN loss and selectivity; directly controls the 99.2% vs 52% outcome.
  • exponential_factor_phi = 10
    Exponent in the probability function of Eq. 3; hand-set, controls gradient behavior of range-loss.
  • rdkit_feature_count = 30
    Number of global RDKit attributes concatenated with Morgan fingerprints; selected by unspecified redundancy removal, not justified quantitatively.
  • morgan_radius = 2
    Morgan fingerprint radius used for descriptor; set in Section V, though not systematically tuned.
  • desired_classifier_range = [0.5, 1]
    Target interval for odorant probability in range-loss; hand-set threshold for binary selectivity.
assumptions (4)
  • domain assumption Morgan fingerprints plus 30 RDKit attributes contain enough information to reconstruct and generate valid molecular SMILES strings.
    The whole transformer embedding pipeline assumes this descriptor is a sufficient representation; only reconversion accuracy on known molecules is offered as support (Section II.A, IV.A).
  • domain assumption The odorant classifier O-1's probabilistic output is a valid oracle for true odorancy.
    Range-loss optimizes O-1's BCE (Section II.C) and Table II evaluates GAN outputs using this same classifier; no independent olfactory validation is provided.
  • domain assumption DeepOlf labels are correct and representative for odorant/non-odorant classification.
    The classifier is trained on DeepOlf's roughly 3,800 labels, and all downstream selectivity claims inherit this label quality.
  • domain assumption ZINC compounds represent the relevant chemical distribution for odorant generation.
    Transformer pretraining and GAN training use ZINC; if ZINC is not representative of odorant chemical space, generated molecules may not transfer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improved Molecular Generation through Attribute-Driven Integrative Embeddings and GAN Selectivity." pith.science (2026). https://pith.science/paper/EXRXUYE4

@misc{pith2026250419040,
  author       = {Pith},
  title        = {Pith review of: Improved Molecular Generation through Attribute-Driven Integrative Embeddings and GAN Selectivity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EXRXUYE4}},
  note         = {Machine review of arXiv:2504.19040}
}
read the original abstract

The growing demand for molecules with tailored properties in fields such as drug discovery and chemical engineering has driven advancements in computational methods for molecular design. Machine learning-based approaches for de-novo molecular generation have recently garnered significant attention. This paper introduces a transformer-based vector embedding generator combined with a modified Generative Adversarial Network (GAN) to generate molecules with desired properties. The embedding generator utilizes a novel molecular descriptor, integrating Morgan fingerprints with global molecular attributes, enabling the transformer to capture local functional groups and broader molecular characteristics. Modifying the GAN generator loss function ensures the generation of molecules with specific desired properties. The transformer achieves a reconversion accuracy of 94% while translating molecular descriptors back to SMILES strings, validating the utility of the proposed embeddings for generative tasks. The approach is validated by generating novel odorant molecules using a labeled dataset of odorant and non-odorant compounds. With the modified range-loss function, the GAN exclusively generates odorant molecules. This work underscores the potential of combining novel vector embeddings with transformers and modified GAN architectures to accelerate the discovery of tailored molecules, offering a robust tool for diverse molecular design applications.

Figures

Figures reproduced from arXiv: 2504.19040 by the authors.

Figure 2
Figure 2. Range-constrained GAN Architecture 1) Range-Loss: Modified Loss Function to Generate Molecules with Specific Traits: The loss function for the generator is modified to generate more odorant molecules selectively. This approach is adapted from Nobari et al. [25], where this loss function was first proposed. The Generator objective is modified as Equation 1 min G (−1 · Ez∼Pz [D(G(z)] + λ1 · Lrng) (1) The term Ez∼Pz [D… view at source ↗
Figure 3
Figure 3. Values of Range-Loss GAN with 15k elements of ZINC Dataset - [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 22 canonical work pages

  1. [24]

    Arjovsky, M., Chintala, S., & Bottou, L. (2017). Wasserstein GAN. Proceedings of the 34th International Conference on Machine Learning (ICML), 70, 214–223

    work page 2017

  2. [25]

    Heyrani Nobari, Amin, Wei Chen, and Faez Ahmed. ”Range-GAN: Range-constrained generative adversarial network for conditioned design synthesis.” International Design Engineering Technical Conferences and Computers and Information in Engineering Conference. V ol. 85390. American Society of Mechanical Engineers, 2021

    work page 2021

  3. [1]

    High Throughput Screening: A Comprehen- sive Review of Technologies and Applications

    Sittampalam, G. S., et al. “High Throughput Screening: A Comprehen- sive Review of Technologies and Applications.” Advances in Experi- mental Medicine and Biology, vol. 1033, 2017, pp. 45–80

    work page 2017

  4. [2]

    Yang, Jing Lu, and Yee Sin Ang

    Chuin Wei Tan, Linqiang Xu, Chen Chen Er, Siang-Piao Chai, Boris Kozinsky,Hui Ying Yang, Shengyuan A. Yang, Jing Lu, and Yee Sin Ang. Toward sustainable ultrawide bandgap van der waals materi- als: An ab initio screening effort. Advanced Functional Materials, 34(11):2308679

    work page

  5. [3]

    J Comput Aided Mol Des 27(8):675–679

    Polishchuk PG, Madzhidov TI, Varnek A (2013) Estimation of the size of drug-like chemical space based on gdb-17 data. J Comput Aided Mol Des 27(8):675–679

    work page 2013

  6. [4]

    Nat Chem 8(6):531

    Rodrigues T, Reker D, Schneider P, Schneider G (2016) Counting on natural products for drug design. Nat Chem 8(6):531

    work page 2016

  7. [5]

    ”Automatic chemical design using a data-driven continuous representation of molecules.” ACS central science 4.2 (2018): 268-276

    G ´omez-Bombarelli, Rafael, et al. ”Automatic chemical design using a data-driven continuous representation of molecules.” ACS central science 4.2 (2018): 268-276

    work page 2018

  8. [6]

    MolGAN: An implicit generative model for small molecular graphs

    De Cao, Nicola, and Thomas Kipf. “MolGAN: An implicit generative model for small molecular graphs.” arXiv preprint arXiv:1805.11973 (2018)

    arXiv 2018

Show all 27 references
  1. [7]

    REINVENT 2.0: An AI Tool for De Novo Drug Design

    Blaschke, Thomas et al. “REINVENT 2.0: An AI Tool for De Novo Drug Design.” Journal of chemical information and modeling vol. 60,12 (2020): 5918-5922. doi:10.1021/acs.jcim.0c00915

    2020 doi

  2. [8]

    Abbasi, M., Santos, B.P., Pereira, T.C. et al. Designing optimized drug candidates with Generative Adversarial Network. J Cheminform 14, 40 (2022)

    2022

  3. [10]

    Gomez, Łukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin

  4. [11]

    Smiles-bert: large scale unsupervised pre-training for molecular property prediction

    Wang, S.; Guo, Y .; Wang, Y .; Sun, H.; Huang, J. Smiles-bert: large scale unsupervised pre-training for molecular property prediction. Pro- ceedings of the 10th ACM international conference on bioinformatics, computational biology and health informatics 2019; pp 429–436

    2019

  5. [12]

    Irwin, R.; Dimitriadis, S.; He, J.; Bjerrum, E. J. Chemformer: a pre- trained transformer for computational chemistry. Machine Learning: Science and Technology 2022, 3, 015022

    2022

  6. [13]

    Large scale chemical language representations capture molecular structure and properties

    Ross, J.; Belgodere, B.; Chenthamarakshan, V .; Padhi, I.; Mroueh, Y .; Das, P. Large scale chemical language representations capture molecular structure and properties. Nature Machine Intelligence 2022, 4, 1256–1264

    2022

  7. [14]

    Knowledge-based BERT: a method to extract molecular features like computational chemists

    Wu, Z.; Jiang, D.; Wang, J.; Zhang, X.; Du, H.; Pan, L.; Hsieh, C.-Y .; Cao, D.; Hou, T. Knowledge-based BERT: a method to extract molecular features like computational chemists. Briefings in Bioinformatics 2022, 23, bbac131

    2022

  8. [15]

    Devlin, J., Chang, M., Lee, K., & Toutanova, K. (2019). BERT: Pre- training of Deep Bidirectional Transformers for Language Understand- ing. North American Chapter of the Association for Computational Linguistics

    2019

  9. [16]

    & Hamaekers, J

    Dobberstein, N., Maass, A. & Hamaekers, J. Llamol: a dynamic multi- conditional generative transformer for de novo molecular design. J Cheminform 16, 73 (2024). https://doi.org/10.1186/s13321-024-00863- 8

    2024 doi

  10. [17]

    Mazuz, E., Shtar, G., Shapira, B. et al. Molecule generation using transformers and policy gradient reinforcement learning. Sci Rep 13, 8799 (2023). https://doi.org/10.1038/s41598-023-35648

    2023 doi

  11. [18]

    Wen, N., Liu, G., Zhang, J. et al. A fingerprints based molecular property prediction method using the BERT model. J Cheminform 14, 71 (2022)

    2022

  12. [19]

    ”Convolutional Neural Network Model Based on 2D Fingerprint for Bioactivity Prediction.” International Journal of Molecular Sciences, vol

    Hentabli, H., et al. ”Convolutional Neural Network Model Based on 2D Fingerprint for Bioactivity Prediction.” International Journal of Molecular Sciences, vol. 23, no. 21, 2022, https://doi.org/10.3390/ijms232113230

    2022 doi

  13. [20]

    Rogers, D., & Hahn, M. (2010). Extended-Connectivity fingerprints. Journal of Chemical Information and Modeling, 50(5), 742–754. https://doi.org/10.1021/ci100050t

    2010 doi

  14. [21]

    Correction: Reconstruction of lossless molecular representations from fingerprints

    Umit V Ucak, Islambek Ashyrmamatov, and Juyong Lee. Correction: Reconstruction of lossless molecular representations from fingerprints. J. Cheminform., 15(1):68, July 2023

    2023

  15. [22]

    RDKit: Open-source cheminformatics; http://www.rdkit.org

  16. [23]

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y . (2014). Generative adver- sarial nets. Advances in Neural Information Processing Systems, 27, 2672–2680

    2014

  17. [26]

    ZINC–a free database of commer- cially available compounds for virtual screening

    John J Irwin and Brian K Shoichet. ZINC–a free database of commer- cially available compounds for virtual screening. J. Chem. Inf. Model., 45(1):177–182, January 2005

    2005

  18. [27]

    Sharma, Anju, et al. ”DeepOlf: Deep Neural Network Based Architec- ture for Predicting Odorants and Their Interacting Olfactory Receptors.” IEEE/ACM Transactions on Computational Biology and Bioinformatics, vol. 19, no. 1, Jan. 2022, pp. 418–428

    2022

  19. [2017]

    In Proceedings of the 31st In- ternational Conference on Neural Information Processing Systems (NIPS’17)

    Attention is all you need. In Proceedings of the 31st In- ternational Conference on Neural Information Processing Systems (NIPS’17). Curran Associates Inc., Red Hook, NY , USA, 6000–6010. https://doi.org/10.1021/ci00057a005

    doi

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.