Pith. sign in

REVIEW 3 major objections 6 minor 63 references

A Collaborative Framework Integrating Large Language Model and Chemical Fragment Space: Mutual Inspiration for Lead Design

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read AutoLeadDesign couples an LLM with docking-scored fragment feedback to outperform four baselines and generate expert-competitive leads with new binding modes.

desk verdict A genuinely new closed-loop LLM-plus-fragment design framework that optimizes docking scores impressively, but the validation is entirely self-referential and the expert-level claims are not supported. read the letter →

arxiv 2507.13580 v2 pith:XYVQGK4L submitted 2025-07-17 q-bio.BM cs.LG

Hao Tuo , Yan Li , Xuanning Hu , Haishi Zhao , Xueyan Liu , Bo Yang This is my paper
classification q-bio.BMcs.LG
keywords denovodrugdesignlargelanguagemodelfragment-basedchemicalspaceexplorationdockingaffinityscoringleadcompoundPRMT5SARS-CoV-2PLpro
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

AutoLeadDesign is an iterative loop that uses a large language model to propose new molecules and uses docking affinity to decide which chemical fragments are worth keeping. The paper aims to establish that this loop lets the LLM's latent chemical knowledge be guided by target-specific fragment scores, so that even without starting from known bioactive ligands it can explore vast chemical space and design high-affinity lead compounds. In benchmarks over ten binding pockets, the authors report top-1/10/100 docking scores above those of REINVENT, ChemGE, RGA, and LMLF, and in case studies on PRMT5 and SARS-CoV-2 PLpro they report de novo compounds with lower docking energies than expert-designed inhibitors and new interaction patterns such as an additional hydrogen bond to GLU444. If correct, this turns an LLM into a practical lead-discovery engine for targets where no active compounds are known, including newly emerging pathogens and rare-disease proteins.

What carries the argument

The load-bearing mechanism is the iterative two-way mapping between molecule space and fragment space. On the descending side, BRICS decomposes each molecule into fragments with at least two heavy atoms, and equation (1) scores a fragment as the average smina docking score over all compounds containing it; the top-k fragments form the filtered library and are sampled with probability proportional to their score via equation (2). On the ascending side, the sampled fragments are placed into a fixed prompt, and the LLM generates new SMILES that must contain at least one of them; the new molecules are docked with smina to become labels for the next iteration. The fragment importance score is the identity that carries the argument: it converts raw docking outputs into a rankable, interpretable signal that simultaneously updates the library and defines the sampling distribution, which is what lets the LLM's prior chemical knowledge be redirected by target-specific feedback.

What would settle it

Synthesize or purchase the top-ranked compounds AutoLeadDesign generates for PRMT5 and PLpro and measure their IC50 or binding affinity; if the smina docking order does not roughly match the experimental order, the central claim that the loop discovers high-affinity leads is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that the mutual inspiration between an LLM and an affinity-scored fragment space yields a de novo lead-design strategy that outperforms existing baseline methods. Each generation decomposes the current compound library with BRICS, scores each fragment by the average smina docking score of all compounds containing it, samples top fragments proportionally to that score, and prompts DeepSeek-v3 to generate a SMILES containing at least one sampled fragment. The authors report that this cycle beats REINVENT, ChemGE, RGA, and LMLF on top-1/10/100 docking scores across ten pockets, and that traced design decisions reproduce the expert moves of fragment linking, merging, and growing. The case studies are offered as evidence of both affinity and mechanistic novelty: against PRMT5, ADD001 retains the SAM-competitive contacts of LLY while adding a hydrogen bond to GLU444; against PLpro, top compounds engage the required ASP164/TYR268 interactions with lower docking energies than GRL0617 and Jun12682. The paper concludes that AutoLeadDesign is an efficient and broadly applicable approach to lead generation, especially when known bioactive ligands are unavailable.

Load-bearing premise

The entire pipeline treats smina docking scores as a valid proxy for true binding affinity, so if those scores are biased or uncorrelated with experimental binding for these targets, the claimed high-affinity leads and mechanism-validated inhibition do not follow.

Editorial extensions

If this is right

  • A new protein target requires only a random small-molecule library and a pocket structure, removing the usual dependence on known bioactive ligands for generative lead design.
  • Designed compounds can carry genuinely new binding modes, such as the extra GLU444 contact in ADD001, so the method can propose mechanistically distinct inhibitors rather than only optimising known scaffolds.
  • Because the loop works with explicit fragments and affinity scores, the design process is auditable and can be traced to standard fragment-based drug design strategies, aiding medicinal-chemistry review.
  • Low sensitivity to random initialization, as seen in the PLpro campaign with a 0.16 standard deviation in docking energy, indicates the method does not rely on a carefully curated starting library.
  • The framework depends only on a docking function and an LLM, so the same protocol transfers directly to other targets and can in principle be adapted to other molecular properties.

Reading between the lines

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

  • A decisive stress test is to replace smina with a more rigorous free-energy estimator and check whether the selected fragments and final rankings survive; if they shift substantially, the method's chemistry is partly an artifact of the scoring function.
  • The claim that the LLM intrinsically performs fragment-based design rests on post-hoc traces; a cleaner attribution would compare against a non-LLM fragment assembler or an LLM prompted with irrelevant chemistry to quantify the LLM's contribution.
  • The fragment importance score could absorb additional objectives, such as synthetic accessibility or selectivity, turning AutoLeadDesign into a general fragment-based multi-objective optimizer rather than an affinity-only one.
  • Docking energies alone cannot settle 'expert-competitive' potency; experimental IC50 or binding-affinity measurements on a few designed compounds are the decisive evidence needed to confirm the paper's practical promise.
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.

Referee Report

3 major / 6 minor

Summary. The paper proposes AutoLeadDesign, an iterative framework that combines large language models (LLMs) with chemical fragment space for de novo lead compound design. In each iteration, molecules in the current library are decomposed into fragments via BRICS, fragments are scored by the average smina docking score of all compounds containing them (Eq. 1), the top fragments are sampled with weights derived from those scores (Eq. 2), and an LLM (DeepSeek-v3) is prompted to generate new molecules containing at least one sampled fragment. The new molecules are docked with smina and added to the library. The authors benchmark AutoLeadDesign against REINVENT, ChemGE, RGA, and LMLF on 10 CrossDocked2020 targets under random and prior initialization, and present case studies for PRMT5 and SARS-CoV-2 PLpro. They report improved top-1/10/100 smina docking scores over baselines, analyze generated molecules' binding modes and physicochemical properties, and argue that the LLM's decisions follow fragment linking, merging, and growing strategies. The paper claims expert-competitive design efficacy and mechanism-validated inhibitory patterns for the designed compounds.

Significance. If the claims are substantiated, AutoLeadDesign would be a valuable contribution to de novo drug design: it provides a closed-loop framework that combines the chemical knowledge embedded in LLMs with fragment-based exploration, and it is demonstrated on two clinically relevant targets without relying on known bioactive ligands. The paper includes a benchmark against several established methods, a clear description of the algorithm, and a public code/data repository (Section 5). The analysis tracing fragment linking, merging, and growing behaviors (Sections 2.4-2.5) is also a useful conceptual contribution that connects the proposed pipeline to established fragment-based drug design principles. However, the evaluation is almost entirely based on smina docking scores, which are both the optimization target and the success metric, so the stronger claims of high-affinity leads and mechanism-validated inhibition require independent validation or substantially more careful framing.

major comments (3)
  1. [Section 2.1, Table 1] The claim that AutoLeadDesign 'demonstrates statistically superior performance' is not supported by the reported data. Table 1 presents only mean top-1/10/100 smina docking scores, with no standard deviations, confidence intervals, or significance tests across the 10 targets or across multiple runs. The differences between AutoLeadDesign and the best baseline are small in some settings (e.g., random Top-100: -9.85 vs. -9.82 for ChemGE), so without measures of variability it is not possible to assess whether these differences are meaningful. The authors should report per-target results, error bars, and appropriate statistical tests, or explicitly soften the 'statistically superior' wording.
  2. [Section 3.3, Eqs. (1)-(2) and Section 3.5] The central evaluation endpoint is identical to the optimization objective. Fragment importance (Eq. 1) is defined as the mean smina docking score of compounds containing the fragment, the top fragments are selected and weighted by that score (Eq. 2), and the final benchmark (Table 1) and case-study claims (Figs. 2-5) all use smina docking scores computed with the same pipeline described in Section 3.5. Consequently, the results demonstrate that AutoLeadDesign can optimize smina scores, but the claims that it designs compounds with 'high binding affinity', 'expert-competitive design efficacy', and 'mechanism-validated inhibitory patterns' do not follow without independent validation. No held-out affinity data, alternative scoring function (e.g., a second docking tool, MM-PBSA, or molecular dynamics), or experimental assay is provided. This issue is load-bearing for the manuscript's central claims and should be addressed either by adding independent validation or by substantially rephrasing the claims to be explicitly about smina-score optimization rather than bioactivity.
  3. [Section 2.3, Fig. 5] The comparison with expert-designed inhibitors GRL0617 and Jun12682 is based on smina docking poses and Maestro interaction analysis, and the statement that 'PLP001 have lower binding energy than GRL0617, indicating that PLP001 have stronger bioactivity than GRL0617' conflates predicted docking energy with experimental bioactivity. Without IC50 or Kd measurements, or at least an independent scoring method, this inference is unsupported. The same concern applies to the PRMT5 case study (Section 2.2), where the claim that ADD001 establishes a dual binding mode that 'simultaneously competes with both SAM cofactor and protein substrate binding' is presented as a validated mechanism on the basis of a single docking pose. These claims should be toned down or supported by additional evidence.
minor comments (6)
  1. [Section 2.1] The text mentions 'evaluate these methods and AutoLeadGen', which appears to be a typo for 'AutoLeadDesign'.
  2. [Section 3.5 vs. Section 2.1] The dataset is referred to as 'CrossDocked2020' in Section 2.1 but as 'CrossDocked2022' in Section 3.5; please use a consistent name and verify the correct version.
  3. [Section 3.3 and 3.4] The manuscript does not specify the values of several free parameters that affect the algorithm's behavior, such as the top-k fragment library size, the number of fragments sampled per prompt, the number of optimization generations (20 is stated in Section 2.1 but not in Methods), and the temperature setting rationale. Providing these details would improve reproducibility, especially since the code repository is referenced.
  4. [Section 2.2, Fig. 3] The text describing Fig. 3c-d contains an apparent contradiction: it states that 'ADR001 forms a hydrogen bond with key residues GLU444 indicating high bioactivity while ADR001 only interacts with pharmacologically insignificant residues.' The second occurrence should likely refer to PRR001, not ADR001.
  5. [References] References [9] and [14] are identical (Popov et al., 'Hidden Gem', Molecular informatics 43, e202300207 (2024)); please remove the duplicate.
  6. [Section 3.7] The interaction analysis criteria are described in detail, but the criteria for classifying a binding mode as 'novel' or 'mechanism-validated' in Sections 2.2-2.4 are not defined; please clarify how novelty and validity are determined from the docking poses.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the pipeline is evaluated against external baselines under a common docking-score oracle, and no load-bearing result reduces by construction to its inputs.

full rationale

AutoLeadDesign's fragment-importance score (Eq. 1) is an intermediate optimization signal computed from smina docking scores, and the final benchmark metrics are also smina docking scores. However, this is not a circular reduction: the generated molecules are created by an LLM prompted with fragments, then re-docked and ranked independently, and the comparison is made against external baseline methods (REINVENT, ChemGE, RGA, LMLF) under the same scoring protocol. The paper does not fit a parameter to the reported top-1/10/100 scores and then rename that fit as a prediction. The claims about 'expert-competitive design efficacy' and 'high binding affinity' rest on the validity of smina as an affinity proxy, which is a limitation or correctness risk rather than a circularity. No load-bearing self-citations, imported uniqueness theorems, or ansatz-smuggling citations are present; the FBDD analogy is offered as post-hoc interpretation, not as evidence. The derivation chain is therefore self-contained for what it actually claims computationally.

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

The central claim rests on the docking-score proxy and on the LLM's generation reliability; both are domain assumptions rather than demonstrated facts. The free parameters (k, population size, number of fragments, temperature, iterations) are chosen without sensitivity analysis, so the reported performance could be specific to these settings.

free parameters (5)
  • Top-k fragment library size
    The fragment library retains the top-k highest-affinity fragments, but k is not quantified in the text; the search behavior depends on it.
  • Number of fragments sampled per prompt = 3
    Exactly three sampled fragments are inserted into the prompt; this is a hand-chosen design parameter.
  • Number of optimization generations = 20
    The workflow runs 20 iterations of design and evaluation; no sensitivity analysis is provided.
  • Compounds per generation = 100
    Each generation designs 100 compounds per target; this population size is fixed without justification.
  • LLM temperature = 1.5
    DeepSeek-v3 sampling temperature is set to 1.5, balancing diversity and validity; no ablation is reported.
assumptions (4)
  • domain assumption Docking score (smina) is a valid proxy for binding affinity
    All selection, scoring, and validation in the paper are based on smina docking scores; if docking is inaccurate, the central claims of high affinity and mechanism-validated inhibition fail. Introduced in Section 3.5 and used in Eq. (1).
  • domain assumption BRICS decomposition preserves chemically meaningful fragments for design
    The method assumes that fragments from BRICS are appropriate building blocks for lead design; standard in fragment-based drug design but an assumption.
  • domain assumption The LLM (DeepSeek-v3) can generate valid, diverse, fragment-containing molecules from the prompt
    The framework depends on the LLM's ability to comply with the SMILES prompt and produce useful molecules; no validity rate or diversity statistics are reported.
  • domain assumption The 10 selected CrossDocked2020 pockets are representative
    The benchmark uses 10 randomly selected proteins across five classes, but the selection procedure and reproducibility of the random choice are not described.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Collaborative Framework Integrating Large Language Model and Chemical Fragment Space: Mutual Inspiration for Lead Design." pith.science (2026). https://pith.science/paper/XYVQGK4L

@misc{pith2026250713580,
  author       = {Pith},
  title        = {Pith review of: A Collaborative Framework Integrating Large Language Model and Chemical Fragment Space: Mutual Inspiration for Lead Design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XYVQGK4L}},
  note         = {Machine review of arXiv:2507.13580}
}
read the original abstract

Combinatorial optimization algorithm is essential in computer-aided drug design by progressively exploring chemical space to design lead compounds with high affinity to target protein. However current methods face inherent challenges in integrating domain knowledge, limiting their performance in identifying lead compounds with novel and valid binding mode. Here, we propose AutoLeadDesign, a lead compounds design framework that inspires extensive domain knowledge encoded in large language models with chemical fragments to progressively implement efficient exploration of vast chemical space. The comprehensive experiments indicate that AutoLeadDesign outperforms baseline methods. Significantly, empirical lead design campaigns targeting two clinically relevant targets (PRMT5 and SARS-CoV-2 PLpro) demonstrate AutoLeadDesign's competence in de novo generation of lead compounds achieving expert-competitive design efficacy. Structural analysis further confirms their mechanism-validated inhibitory patterns. By tracing the process of design, we find that AutoLeadDesign shares analogous mechanisms with fragment-based drug design which traditionally rely on the expert decision-making, further revealing why it works. Overall, AutoLeadDesign offers an efficient approach for lead compounds design, suggesting its potential utility in drug design.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 59 canonical work pages

  1. [1]

    & Pichika, M

    Mak, K.-K. & Pichika, M. R. Artificial intelligence in drug development: present status and future prospects. Drug discovery today 24, 773–780 (2019)

    work page 2019

  2. [2]

    Keser¨ u, G. M. & Makara, G. M. The influence of lead discovery strategies on the properties of drug candidates. nature reviews Drug Discovery 8, 203–212 (2009)

    work page 2009

  3. [3]

    Corsello, S. M. et al. The drug repurposing hub: a next-generation drug library and information resource. Nature medicine 23, 405–408 (2017)

    work page 2017

  4. [4]

    Kim, S. et al. Pubchem 2023 update. Nucleic acids research 51, D1373–D1380 (2023)

    work page 2023

  5. [5]

    Irwin, J. J. & Shoichet, B. K. Zinc- a free database of commercially available compounds for virtual screening. Journal of chemical information and modeling 45, 177–182 (2005)

    work page 2005

  6. [6]

    Hajduk, P. J. & Greer, J. A decade of fragment-based drug design: strategic advances and lessons learned. Nature reviews Drug discovery 6, 211–219 (2007)

    work page 2007

  7. [7]

    Resistance-guided discovery of new antibiotics

    Flemming, A. Resistance-guided discovery of new antibiotics. Nature Reviews Drug Discovery 12, 826–826 (2013)

    work page 2013

  8. [8]

    P., Davidson, A

    Beckett, R. P., Davidson, A. H., Drummond, A. H., Huxley, P. & Whittaker, M. Recent advances in matrix metalloproteinase inhibitor research. Drug Discovery Today 1, 16–26 (1996)

    work page 1996

Show all 63 references
  1. [9]

    I., Wellnitz, J., Maxfield, T

    Popov, K. I., Wellnitz, J., Maxfield, T. & Tropsha, A. Hit discovery using docking enriched by generative modeling (hidden gem): A novel computational workflow for accelerated virtual screening of ultra-large 10 chemical libraries. Molecular informatics 43, e202300207 (2024)

    2024

  2. [10]

    Sadybekov, A. A. et al. Synthon-based ligand discovery in virtual libraries of over 11 billion compounds. Nature 601, 452–459 (2022)

    2022

  3. [11]

    Bender, B. J. et al. A practical guide to large-scale docking. Nature protocols 16, 4799–4832 (2021)

    2021

  4. [12]

    & Filikov, A

    Yoon, S., Smellie, A., Hartsough, D. & Filikov, A. Surrogate docking: structure-based virtual screening at high throughput speed. Journal of computer-aided molecular design 19, 483–497 (2005)

    2005

  5. [13]

    Lyu, J. et al. Ultra-large library docking for discovering new chemotypes. Nature 566, 224–229 (2019)

    2019

  6. [14]

    I., Wellnitz, J., Maxfield, T

    Popov, K. I., Wellnitz, J., Maxfield, T. & Tropsha, A. Hit discovery using docking enriched by generative modeling (hidden gem): A novel computational workflow for accelerated virtual screening of ultra-large chemical libraries. Molecular informatics 43, e202300207 (2024)

    2024

  7. [15]

    Discovery of lirafugratinib (rly-4008), a highly selective irreversible small-molecule inhibitor of fgfr2

    Sch¨ onherr, H.et al. Discovery of lirafugratinib (rly-4008), a highly selective irreversible small-molecule inhibitor of fgfr2. Proceedings of the National Academy of Sciences 121, e2317756121 (2024)

    2024

  8. [16]

    Wu, K. et al. Tamgen: drug design with target-aware molecule generation through a chemical language model. Nature Communications 15, 9360 (2024)

    2024

  9. [17]

    Huang, L. et al. A dual diffusion model enables 3d molecule generation and lead optimization based on target pockets. Nature Communications 15, 2657 (2024)

    2024

  10. [18]

    Du, Y. et al. Machine learning-aided generative molecular design. Nature Machine Intelligence 6, 589–604 (2024)

    2024

  11. [19]

    & Jaakkola, T

    Jin, W., Barzilay, R. & Jaakkola, T. Junction tree variational autoencoder for molecular graph generation 2323–2332 (2018)

    2018

  12. [20]

    & Durrant, J

    Spiegel, J. & Durrant, J. Autogrow4: an open-source genetic algorithm for de novo drug design and lead optimization. j cheminform 12 (1): 25 (2020)

    2020

  13. [21]

    & Sun, J

    Fu, T., Gao, W., Coley, C. & Sun, J. Reinforced genetic algorithm for structure-based drug design.Advances in Neural Information Processing Systems 35, 12325–12338 (2022)

    2022

  14. [22]

    & Chen, H

    Olivecrona, M., Blaschke, T., Engkvist, O. & Chen, H. Molecular de-novo design through deep reinforcement learning. Journal of cheminformatics 9, 1–14 (2017)

    2017

  15. [23]

    Yoshikawa, N. et al. Population-based de novo molecule generation, using grammatical evolution.Chemistry Letters 47, 1431–1434 (2018)

    2018

  16. [24]

    Y., Miao, B

    Williams, C. Y., Miao, B. Y., Kornblith, A. E. & Butte, A. J. Evaluating the use of large language models to provide clinical recommendations in the emergency department. Nature Communications 15, 8236 (2024). 11

    2024

  17. [25]

    Zheng, Z. et al. Large language models for reticular chemistry. Nature Reviews Materials 1–13 (2025)

    2025

  18. [27]

    Castro Nascimento, C. M. & Pimentel, A. S. Do large language models understand chemistry? a conversation with chatgpt. Journal of Chemical Information and Modeling 63, 1649–1655 (2023)

    2023

  19. [28]

    C., Collison, C

    Ramos, M. C., Collison, C. J. & White, A. D. A review of large language models and autonomous agents in chemistry. Chemical Science (2025)

    2025

  20. [29]

    M. Bran, A. et al. Augmenting large language models with chemistry tools. Nature Machine Intelligence 6, 525–535 (2024)

    2024

  21. [30]

    Dagdelen, J. et al. Structured information extraction from scientific text with large language models. Nature Communications 15, 1418 (2024)

    2024

  22. [31]

    Sun, L. et al. Trustllm: Trustworthiness in large language models. arXiv preprint arXiv:2401.05561 3 (2024)

    2024 arXiv

  23. [32]

    Guo, T. et al. What can large language models do in chemistry? a comprehensive benchmark on eight tasks. Advances in Neural Information Processing Systems 36, 59662–59688 (2023)

    2023

  24. [33]

    Zhang, X. et al. Artificial intelligence for science in quantum, atomistic, and continuum systems. arXiv preprint arXiv:2307.08423 (2023)

    2023 arXiv

  25. [34]

    Wu, Z. et al. Leveraging language model for advanced multiproperty molecular optimization via prompt engineering. Nature Machine Intelligence 1–11 (2024)

    2024

  26. [35]

    & Li, J.-R

    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 64, 4958–4965 (2024)

    2024

  27. [36]

    Yang, F. et al. Empower large language model to perform better on industrial domain-specific question answering. arXiv preprint arXiv:2305.11541 (2023)

    2023 arXiv

  28. [37]

    Jones, C. R. & Bergen, B. K. Large language models pass the turing test. arXiv preprint arXiv:2503.23674 (2025)

    2025 arXiv

  29. [38]

    Brahmavar, S. B. et al. Generating novel leads for drug discovery using llms with logical feedback. Proceedings of the AAAI Conference on Artificial Intelligence 38, 21–29 (2024)

    2024

  30. [39]

    Francoeur, P. G. et al. Three-dimensional convolutional neural networks and a cross-docked data set for structure-based drug design. Journal of chemical information and modeling 60, 4200–4215 (2020)

    2020

  31. [40]

    & Yuan, Y

    Wang, Y., Hu, W. & Yuan, Y. Protein arginine methyltransferase 5 (prmt5) as an anticancer target and its inhibitor discovery. Journal of medicinal chemistry 61, 9429–9441 (2018). 12

    2018

  32. [41]

    & Wang, Y

    Zheng, J., Li, B., Wu, Y., Wu, X. & Wang, Y. Targeting arginine methyltransferase prmt5 for cancer therapy: updated progress and novel strategies. Journal of Medicinal Chemistry 66, 8407–8427 (2023)

    2023

  33. [42]

    Wu, Q., Schapira, M., Arrowsmith, C. H. & Barsyte-Lovejoy, D. Protein arginine methylation: from enigmatic functions to therapeutic targeting. Nature reviews Drug discovery 20, 509–530 (2021)

    2021

  34. [43]

    & Falchook, G

    Feustel, K. & Falchook, G. S. Protein arginine methyltransferase 5 (prmt5) inhibitors in oncology clinical trials: a review. Journal of Immunotherapy and Precision Oncology 5, 58–67 (2022)

    2022

  35. [44]

    Bonday, Z. Q. et al. Lly-283, a potent and selective inhibitor of arginine methyltransferase 5, prmt5, with antitumor activity. ACS medicinal chemistry letters 9, 612–617 (2018)

    2018

  36. [45]

    Brehmer, D. et al. Discovery and pharmacological characterization of jnj-64619178, a novel small-molecule inhibitor of prmt5 with potent antitumor activity. Molecular cancer therapeutics 20, 2317–2328 (2021)

    2021

  37. [46]

    Klemm, T. et al. Mechanism and inhibition of the papain-like protease, plpro, of sars-cov-2. The EMBO journal 39, e106275 (2020)

    2020

  38. [47]

    A., Banerjee, S., Ghosh, K., Gayen, S

    Amin, S. A., Banerjee, S., Ghosh, K., Gayen, S. & Jha, T. Protease targeted covid-19 drug discovery and its challenges: Insight into viral main protease (mpro) and papain-like protease (plpro) inhibitors. Bioorganic & medicinal chemistry 29, 115860 (2021)

    2021

  39. [48]

    Rut, W. et al. Activity profiling and structures of inhibitor-bound sars-cov-2-plpro protease provides a framework for anti-covid-19 drug design. biorxiv 2020–04 (2020)

    2020

  40. [49]

    Shen, Z. et al. Design of sars-cov-2 plpro inhibitors for covid-19 antiviral therapy leveraging binding cooperativity. Journal of medicinal chemistry 65, 2940–2955 (2021)

    2021

  41. [50]

    Yan, S. & Wu, G. Spatial and temporal roles of sars-cov plpro—a snapshot. The F ASEB Journal 35, e21197 (2021)

    2021

  42. [51]

    McClain, C. B. & Vabret, N. Sars-cov-2: the many pros of targeting plpro. Signal Transduction and Targeted Therapy 5, 223 (2020)

    2020

  43. [52]

    Puhl, A. C. et al. Discovery of plpro and mpro inhibitors for sars-cov-2. ACS omega 8, 22603–22612 (2023)

    2023

  44. [53]

    L., Zhang, A., Tatineni, M., Miller, M

    Kouznetsova, V. L., Zhang, A., Tatineni, M., Miller, M. A. & Tsigelny, I. F. Potential covid-19 papain-like protease plpro inhibitors: repurposing fda-approved drugs. PeerJ 8, e9965 (2020)

    2020

  45. [54]

    Ershov, P. et al. Sars-cov-2 coronavirus papain-like protease plpro as an antiviral target for inhibitors of active site and protein–protein interactions. Biophysics 67, 902–912 (2022)

    2022

  46. [55]

    Shin, D. et al. Papain-like protease regulates sars-cov-2 viral spread and innate immunity. Nature 587, 657–662 (2020). 13

    2020

  47. [56]

    Ratia, K. et al. A noncovalent class of papain-like protease/deubiquitinase inhibitors blocks sars virus replication. Proceedings of the National Academy of Sciences 105, 16119–16124 (2008)

    2008

  48. [57]

    Jadhav, P. et al. Structure-based design of sars-cov-2 papain-like protease inhibitors. European journal of medicinal chemistry 264, 116011 (2024)

    2024

  49. [58]

    Von Delft, A. et al. Accelerating antiviral drug discovery: lessons from covid-19. Nature Reviews Drug Discovery 22, 585–603 (2023)

    2023

  50. [59]

    & Cheng, W

    Yan, W., Zheng, Y., Zeng, X., He, B. & Cheng, W. Structural biology of sars-cov-2: open the door for novel therapies. Signal Transduction and Targeted Therapy 7, 26 (2022)

    2022

  51. [60]

    Narayanan, A. et al. Identification of sars-cov-2 inhibitors targeting mpro and plpro using in-cell-protease assay. Communications Biology 5, 169 (2022)

    2022

  52. [61]

    Garnsey, M. R. et al. Discovery of sars-cov-2 papain-like protease (plpro) inhibitors with efficacy in a murine infection model. Science Advances 10, eado4288 (2024)

    2024

  53. [62]

    Tan, B. et al. Design of a sars-cov-2 papain-like protease inhibitor with antiviral efficacy in a mouse model. Science 383, 1434–1440 (2024)

    2024

  54. [63]

    R., Baumgartner, M

    Koes, D. R., Baumgartner, M. P. & Camacho, C. J. Lessons learned in empirical scoring with smina from the csar 2011 benchmarking exercise. Journal of chemical information and modeling 53, 1893–1904 (2013)

    2013

  55. [64]

    & Coley, C

    Gao, W., Fu, T., Sun, J. & Coley, C. Sample efficiency matters: a benchmark for practical molecular optimization. Advances in neural information processing systems 35, 21342–21357 (2022). 14 Fig. 1 The workflow of AutoLeadDesign. a, The molecule space. The compound library con...

    2022

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.