Pith. sign in

REVIEW 3 major objections 3 minor 42 references

DrugGen 2: A disease-aware language model for enhancing drug discovery

T0 review · 3 major / 3 minor · reviewed 2026-07-10 · grok-4.5

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

Pith's one-line read DrugGen-2 generates small molecules conditioned on both disease ontology and target sequence, outperforming prior models on diabetic nephropathy targets.

desk verdict Solid open-source incremental step that conditions LM generation on MeSH+sequence and beats its own baselines on the metrics it optimizes; the PLAPT circularity is real but does not erase the rest of the work. read the letter →

arxiv 2607.08404 v1 pith:5IWOWTUQ submitted 2026-07-09 q-bio.QM cs.AIcs.LG

Ali Motahharynia , Mohammadreza Ghaffarzadeh-Esfahani , Mahsa Sheikholeslami , Navid Mazrouei , Matin Irajpour , Yousof Gheisari , Hajar Sirous This is my paper
classification q-bio.QMcs.AIcs.LG
keywords DrugdesignrepositioningLargelanguagemodelReinforcementlearningDisease-conditionedgenerationMeSHontologySMILESBindingaffinity
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

Most generative drug models condition only on a protein target or generic properties, even though the same target can behave differently depending on the disease. DrugGen-2 closes that gap by taking a disease MeSH hierarchy plus a protein sequence and emitting SMILES strings tailored to that disease–target pair. The authors fine-tune a GPT-2 model on approved disease–target–drug triples, then refine it with group relative policy optimization using rewards for validity, novelty, diversity, and predicted binding affinity. On five diabetic-nephropathy targets the model produces more unique valid molecules, closer structural analogues of approved drugs, and higher predicted affinities than two earlier baselines; docking further flags candidates that score better than reference drugs such as enalapril. The result is a practical route to disease-aware de novo design and repurposing that treats context as an explicit input rather than an afterthought.

What carries the argument

Two-stage training of a GPT-2 base: supervised fine-tuning on curated MeSH–sequence–SMILES strings, followed by group relative policy optimization (GRPO) driven by three batch-level rewards (PLAPT binding affinity plus validity, binary novelty versus approved drugs, and intra-batch uniqueness).

What would settle it

Synthesize the top ACE and PPARγ ligands reported in the docking table, measure their experimental Kd or IC50 against the same proteins, and check whether the measured affinities rank in the same order and exceed the reference drugs as the in-silico scores predict.

Watch

Extended reading notes

Core claim

Conditioning a generative language model on both disease ontology (MeSH DAG) and target protein sequence, then refining it with group-relative policy optimization under validity, novelty, diversity and affinity rewards, yields molecules that are more unique, more similar to approved drugs, and higher-scoring on predicted binding affinity than models conditioned only on the target.

Load-bearing premise

The claim that higher PLAPT affinity scores and Glide docking scores mean better real-world drug candidates rests on those computational proxies being accurate enough substitutes for experimental binding and therapeutic effect.

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 / 3 minor

Summary. DrugGen-2 is a GPT-2-based generative model fine-tuned first by supervised fine-tuning and then by group relative policy optimization (GRPO) to produce SMILES conditioned on both MeSH disease DAG hierarchies and target protein sequences. Training uses a curated set of approved disease–target–drug triples and three rewards (PLAPT-predicted binding affinity with invalid-structure zeroing, binary novelty vs. approved drugs, and batch-level uniqueness). On five diabetic-nephropathy targets the model is reported to generate more unique and valid molecules, higher Tanimoto similarity to approved drugs, and higher PLAPT pKd scores than DrugGPT and DrugGen; limited Glide XP docking on ACE and PPARγ is offered as corroboration.

Significance. Conditioning molecular generation on disease ontology as well as target sequence is a genuine and under-explored direction that could improve context-aware design and repurposing. The public release of the MeSH–sequence–SMILES dataset, model checkpoints, code, and interactive demo is a clear strength that supports reproducibility. If the performance gains can be shown to hold under independent affinity estimators and experimental validation, the work would constitute a useful advance for disease-aware generative chemistry.

major comments (3)
  1. [§2.5 / §4.3.4 / §4.4.4] Sections 2.5, 4.3.4 and 4.4.4: the primary binding-affinity metric (PLAPT pKd) is identical to the dominant GRPO reward. Consequently the large, statistically significant gains over DrugGPT/DrugGen largely measure successful optimization of that surrogate rather than independent evidence of better binders. An orthogonal affinity method (e.g., AutoDock Vina, MM-GBSA, or a held-out experimental set) is required before the central claim of improved binding can be accepted.
  2. [§2.6 / Table 1] Section 2.6 and Table 1: docking is only partial corroboration. Three ACE ligands beat enalapril, yet none of the PPARγ ligands beat the reference drugs; re-docking RMSD for the ACE co-crystal is 4.87 Å, indicating unreliable pose recovery for that target. The docking results therefore cannot rescue the circular PLAPT evaluation.
  3. [§4.2] Section 4.2: protein sequences are truncated at 768 tokens. For multi-domain or allosteric targets this may discard functionally critical regions, yet no ablation or sensitivity analysis is provided. The effect of truncation on both generation quality and PLAPT scores should be quantified.
minor comments (3)
  1. [§2 / Fig. 2] Figure 2 panels and Supporting Information tables are referenced extensively but the main text does not always state the exact sample sizes or multiple-testing corrections used for each panel; a short methods note would improve clarity.
  2. [§4.3.4] The binary novelty reward (Section 4.3.4) treats any molecule already present in the approved set as zero; a continuous similarity-based novelty term might better encourage scaffold hopping while still avoiding exact duplicates.
  3. [Throughout] Typographical inconsistencies appear (e.g., “resoectively”, mixed en-dashes, occasional missing spaces around MeSH identifiers). A careful proof-read is needed.

Circularity Check

2 steps flagged · score 6.0 of 10

PLAPT pKd is both the dominant GRPO reward and the primary reported affinity metric, so the headline binding-affinity superiority is partly by construction.

  1. fitted input called prediction [§4.3.4 (reward) + §2.5 / §4.4.4 (evaluation)]
    "•Binding affinity: To measure the binding affinity, we used PLAPT... If the SMILES is valid, the function extracts the protein sequence... and... is then passed to the PLAPT model, which returns a predicted binding affinity score defined as the negative logarithm (base 10) of the dissociation constant (Kd). ... The same set of molecules generated during the validity assessment was used to evaluate the binding affinities of the compounds produced by DrugGen-2, DrugGen, and DrugGPT. ... DrugGen-2 consistently outperformed DrugGPT and DrugGen across all five DN-associated targets ... Median affin"

    GRPO explicitly maximizes the PLAPT pKd reward (invalid SMILES get 0; valid ones receive PLAPT’s –log10(Kd)). The identical PLAPT scores are then reported as the key evaluation metric showing superiority. The affinity gains are therefore the direct, expected outcome of the optimization objective rather than an independent prediction or first-principles result.

  2. fitted input called prediction [§4.3.4 (novelty/diversity rewards) + §2.2 (unique-generation claim)]
    "•Similarity to approved molecules: ... A binary reward was assigned such that molecules already present in the dataset received a reward of zero, whereas novel molecules ... received a reward of one. •Avoidance of repeated molecules within a batch: ... binary reward of one if it was unique within the batch, or zero if it appeared multiple times ... DrugGen-2 consistently achieved the highest number of unique generations, ranging from 409 [406–427] to 444 [443–448] compared to DrugGen (50 [41–87]) and DrugGPT (219 [217–254])"

    The GRPO rewards directly incentivize novelty relative to the approved-drug set and uniqueness inside each training batch. The subsequent claim of superior unique-generation capacity therefore partly reflects the same objectives that were optimized, although the evaluation protocol (500 unique molecules per target) is not identical to the batch-level reward and retains some independent content.

full rationale

The central performance claim (higher predicted binding affinities than DrugGPT/DrugGen on five DN targets) rests on PLAPT scores. The identical PLAPT model supplies the binding-affinity component of the GRPO reward that was used to train DrugGen-2. Consequently the numerical gains on that metric largely measure successful optimization against the training oracle rather than independent evidence of better binders. Novelty/diversity/validity rewards produce milder circularity on the uniqueness and validity metrics. Docking (Glide XP) supplies partial orthogonal support for a subset of ACE ligands but is not the primary metric and fails to beat references on PPARγ. No other independent affinity method or experimental data is reported. The derivation of the affinity result therefore reduces, by the paper’s own reward and evaluation definitions, to the training objective; other claims (unique generation capacity, structural similarity) retain independent content. Score 6 reflects this partial, load-bearing circularity on the strongest quantitative claim.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central performance claims rest on a handful of free training choices, standard domain assumptions about cheminformatics tools, and the unproven transferability of PLAPT and Glide scores to real pharmacology. No new physical entities are postulated.

free parameters (4)
  • SFT learning rate / epochs / batch size
    Chosen as 5e-4, 10 epochs (model at epoch 6 kept), batch 8; directly control how well the model memorizes the 13 908 training strings.
  • GRPO learning rate / group size / temperature / β KL weight
    5e-6, G=40, T=1.0 and unspecified β; these govern the strength of the multi-objective reward optimization that produces the reported gains.
  • Sequence truncation length 768 tokens
    Hard cut-off applied to all protein sequences; may discard distal domains that affect binding.
  • PLAPT reward scale and invalid-structure zeroing
    Invalid SMILES receive reward 0; valid ones receive raw PLAPT pKd. The relative weighting among the three rewards is not stated and therefore free.
assumptions (5)
  • domain assumption PLAPT pKd is a reliable surrogate for experimental binding affinity
    Used both as training reward and as primary evaluation metric (Sections 2.5, 4.3.4).
  • ad hoc to paper MeSH DAG hierarchies encode disease-specific target behavior sufficiently for generation
    Core modeling choice that distinguishes DrugGen-2 from prior target-only models; never independently validated.
  • domain assumption RDKit valence and atom-count rules define chemical validity
    Standard cheminformatics filter applied during training and evaluation.
  • domain assumption Tanimoto similarity on Morgan fingerprints measures drug-likeness
    Used to claim greater structural similarity to approved drugs.
  • domain assumption Glide XP scores with 40 Å boxes and OPLS4 preparation rank true binders
    Docking protocol whose reliability is only partially checked by re-docking (RMSD 1.44 Å vs 4.87 Å).
invented entities (1)
  • DrugGen-2 model (disease-conditioned GPT-2 + GRPO)
    purpose: Generate SMILES given MeSH DAG + protein sequence
    The central artifact; no independent experimental evidence of its outputs is supplied beyond the in-silico metrics used in training.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DrugGen 2: A disease-aware language model for enhancing drug discovery." pith.science (2026). https://pith.science/paper/5IWOWTUQ

@misc{pith2026260708404,
  author       = {Pith},
  title        = {Pith review of: DrugGen 2: A disease-aware language model for enhancing drug discovery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5IWOWTUQ}},
  note         = {Machine review of arXiv:2607.08404}
}
read the original abstract

Current computational approaches for drug design typically focus on generating molecules conditioned on specific targets or general molecular properties, often neglecting the influence of disease context on target behavior and therapeutic outcomes. To address this gap, we introduce DrugGen-2, a novel generative model that designs small molecules conditioned on both disease ontology and target protein sequences. DrugGen-2 was developed by fine-tuning a pre-trained GPT-2 model on a curated dataset of approved drugs linked to their diseases and targets, using a two-step strategy of supervised fine-tuning followed by reinforcement learning via group relative policy optimization (GRPO). This process was guided by reward functions optimizing for chemical validity, novelty, diversity, and high predicted binding affinity. When evaluated on five protein targets relevant to diabetic nephropathy, DrugGen-2 significantly outperformed baseline models (DrugGPT and DrugGen). It demonstrated a superior capacity to generate unique molecules, exhibited greater structural similarity to approved drugs, and achieved improved predicted binding affinities across all targets. Molecular docking analyses further supported these findings, identifying candidate ligands with strong binding potential, including compounds with predicted affinities (-9.917, -9.485, and -9.367) exceeding those of reference drugs such as enalapril for angiotensin-converting enzyme (-8.283). By integrating disease-specific context into molecular generation, DrugGen-2 advances AI-assisted drug discovery, offering a powerful tool for de novo design and drug repurposing that accounts for the complex interplay between diseases and molecular targets.

Figures

Figures reproduced from arXiv: 2607.08404 by the authors.

Figure 1
Figure 1. Schematic representation of DrugGen-2 development and evaluation. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Evaluation of the DrugGen-2 algorithm. (A) [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

  1. [1]

    Advancements in contemporary pharmacological innovation: Mechanistic insights and emerging trends in drug discovery and development.Intell Pharm

    Majumder S, Panigrahi GK. Advancements in contemporary pharmacological innovation: Mechanistic insights and emerging trends in drug discovery and development.Intell Pharm. 2025;3(2):118–26. doi: 10.1016/j.ipha.2024.10.001

    work page doi:10.1016/j.ipha.2024.10.001 2025

  2. [2]

    Phenotypic Drug Discovery: Recent successes, lessons learned and new directions.Nat Rev Drug Discov

    Vincent F, Nueda A, Lee J, Schenone M, Prunotto M, Mercola M. Phenotypic Drug Discovery: Recent successes, lessons learned and new directions.Nat Rev Drug Discov. 2022;21(12):899–914. doi: 10.1038/s41573-022-00472-w

    work page doi:10.1038/s41573-022-00472-w 2022

  3. [3]

    Complementary Approaches to Existing Tar- get Based Drug Discovery for Identifying Novel Drug Targets.Biomedicines

    Vasaikar S, Bhatia P, Bhatia PG, Chu Yaiw K. Complementary Approaches to Existing Tar- get Based Drug Discovery for Identifying Novel Drug Targets.Biomedicines. 2016;4(4):27. doi: 10.3390/biomedicines4040027

    work page doi:10.3390/biomedicines4040027 2016

  4. [4]

    Machine learn- ing in drug design: Use of artificial intelligence to explore the chemical structure–biological activity relationship.WIREs Comput Mol Sci

    Staszak M, Staszak K, Wieszczycka K, Bajek A, Roszkowski K, Tylkowski B. Machine learn- ing in drug design: Use of artificial intelligence to explore the chemical structure–biological activity relationship.WIREs Comput Mol Sci. 2022;12(2):e1568. doi: 10.1002/wcms.1568

    work page doi:10.1002/wcms.1568 2022

  5. [5]

    Generative Deep Learning for de Novo Drug Design—A Chemical Space Odyssey.J Chem Inf Model

    Özçelik R, Brinkmann H, Criscuolo E, Grisoni F. Generative Deep Learning for de Novo Drug Design—A Chemical Space Odyssey.J Chem Inf Model. 65(14):7352–72. doi: 10.1021/acs.jcim.5c00641

    work page doi:10.1021/acs.jcim.5c00641

  6. [6]

    Can large language models understand molecules?BMC Bioinformatics

    Sadeghi S, Bui A, Forooghi A, Lu J, Ngom A. Can large language models understand molecules?BMC Bioinformatics. 2024;25(1):225. doi: 10.1186/s12859-024-05847-x

    work page doi:10.1186/s12859-024-05847-x 2024

  7. [7]

    Mol-LLM: Multimodal Generalist Molecular LLM with Improved Graph Utilization

    Lee C, Ko H, Song Y, Jeong Y, Hormazabal R, Han S, et al. Mol-LLM: Multimodal Gener- alist Molecular LLM with Improved Graph Utilization. arXiv; 2025. [accessed 3 June 2026] Available from:http://arxiv.org/abs/2502.02810

    work page Pith review arXiv 2025

  8. [8]

    J.; Li, J.; Edgar, J

    Wang J, Qin R, Wang M, Fang M, Zhang Y, Zhu Y, et al. Token-Mol 1.0: tokenized drug design with large language models.Nat Commun. 2025;16(1):4416. doi: 10.1038/s41467- 025-59628-y

    work page doi:10.1038/s41467- 2025

Show all 42 references
  1. [9]

    DrugLLM: Open Large Language Model for Few-shot Molecule Generation

    Liu X, Guo Y, Li H, Liu J, Huang S, Ke B, et al. DrugLLM: Open Large Language Model for Few-shot Molecule Generation. arXiv; 2024. [accessed 23 Sept 2025] Available from: http://arxiv.org/abs/2405.06690

    2024 arXiv

  2. [10]

    Transformer-Based Molecular Generative Model for Antiviral Drug Design.J Chem Inf Model

    Mao J, Wang J, Zeb A, Cho K-H, Jin H, Kim J, et al. Transformer-Based Molecular Generative Model for Antiviral Drug Design.J Chem Inf Model. 2023;64(7):2733–45. doi: 10.1021/acs.jcim.3c00536

    2023 doi

  3. [11]

    2024;15(1):9360

    WuK,XiaY,DengP,LiuR,ZhangY,GuoH,etal.TamGen: drugdesignwithtarget-aware molecule generation through a chemical language model.Nat Commun. 2024;15(1):9360. doi: 10.1038/s41467-024-53632-4

    2024 doi

  4. [12]

    DrugGPT: A GPT-based Strategy for De- signing Potential Ligands Targeting Specific Proteins

    Li Y, Gao C, Song X, Wang X, Xu Y, Han S. DrugGPT: A GPT-based Strategy for De- signing Potential Ligands Targeting Specific Proteins. 2023 June 30; [accessed 23 Sept 2025] Available from:https://www.biorxiv.org/content/10.1101/2023.06.29.543848v1

    2023 doi

  5. [13]

    2025;15(1):13445

    SheikholeslamiM,MazroueiN,GheisariY,FasihiA,IrajpourM,MotahharyniaA.DrugGen enhances drug discovery with large language models and reinforcement learning.Sci Rep. 2025;15(1):13445. doi: 10.1038/s41598-025-98629-1

    2025 doi

  6. [14]

    One target, different effects: a comparison of distinct therapeutic antibodies against the same targets.Exp Mol Med

    Shim H. One target, different effects: a comparison of distinct therapeutic antibodies against the same targets.Exp Mol Med. 2011;43(10):539–49. doi: 10.3858/emm.2011.43.10.063 13

    2011 doi

  7. [15]

    Personalized Drug Therapy: Innovative Con- cept Guided With Proteoformics.Mol Cell Proteomics MCP

    Su J, Yang L, Sun Z, Zhan X. Personalized Drug Therapy: Innovative Con- cept Guided With Proteoformics.Mol Cell Proteomics MCP. 2024;23(3):100737. doi: 10.1016/j.mcpro.2024.100737

    2024 doi

  8. [16]

    Interactions between PPAR Gamma and the Canonical Wnt/β-Catenin Pathway in Type 2 Diabetes and Colon Cancer.PPAR Res

    Lecarpentier Y, Claes V, Vallée A, Hébert J-L. Interactions between PPAR Gamma and the Canonical Wnt/β-Catenin Pathway in Type 2 Diabetes and Colon Cancer.PPAR Res. 2017;2017:5879090. doi: 10.1155/2017/5879090

    2017 doi

  9. [17]

    Fine-tuning large language models for domain adaptation: exploration of training strategies, scaling, model merging and synergistic capabilities.Npj Comput Mater

    Lu W, Luu RK, Buehler MJ. Fine-tuning large language models for domain adaptation: exploration of training strategies, scaling, model merging and synergistic capabilities.Npj Comput Mater. 2025;11(1):84. doi: 10.1038/s41524-025-01564-y

    2025 doi

  10. [18]

    DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

    Shao Z, Wang P, Zhu Q, Xu R, Song J, Bi X, et al. DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models. arXiv; 2024. [accessed 17 Aug 2025] Available from:http://arxiv.org/abs/2402.03300

    2024 arXiv

  11. [19]

    DisGeNET: a comprehensive platform integrating information on human disease-associated genes and variants.Nucleic Acids Res

    PiñeroJ,BravoÀ,Queralt-RosinachN,Gutiérrez-SacristánA,Deu-PonsJ,CentenoE,etal. DisGeNET: a comprehensive platform integrating information on human disease-associated genes and variants.Nucleic Acids Res. 2017;45(D1):D833–9. doi: 10.1093/nar/gkw943

    2017 doi

  12. [20]

    DrugTar improves drug- gability prediction by integrating large language models and gene ontologies.Bioinformatics

    Borhani N, Izadi I, Motahharynia A, Sheikholeslami M, Gheisari Y. DrugTar improves drug- gability prediction by integrating large language models and gene ontologies.Bioinformatics. 2025;41(7):btaf360. doi: 10.1093/bioinformatics/btaf360

    2025 doi

  13. [21]

    PLAPT: Protein-Ligand Binding Affinity Predic- tion Using Pretrained Transformers

    Rose T, Monti N, Anand N, Shen T. PLAPT: Protein-Ligand Binding Affinity Predic- tion Using Pretrained Transformers. 2024 Feb 9; [accessed 29 Sept 2025] Available from: https://www.biorxiv.org/content/10.1101/2024.02.08.575577v1

    2024 doi

  14. [22]

    Two Opposing Functions of Angiotensin- Converting Enzyme (ACE) That Links Hypertension, Dementia, and Aging.Int J Mol Sci

    Le D, Brown L, Malik K, Murakami S. Two Opposing Functions of Angiotensin- Converting Enzyme (ACE) That Links Hypertension, Dementia, and Aging.Int J Mol Sci. 2021;22(24):13178. doi: 10.3390/ijms222413178

    2021 doi

  15. [23]

    Angiotensin-converting enzyme degrades Alzheimer amyloid beta-peptide (Aβ); retardsAβaggregation, deposition, fibril formation; and inhibits cytotoxicity.J Biol Chem

    Hu J, Igarashi A, Kamata M, Nakagawa H. Angiotensin-converting enzyme degrades Alzheimer amyloid beta-peptide (Aβ); retardsAβaggregation, deposition, fibril formation; and inhibits cytotoxicity.J Biol Chem. 2001;276(51):47863–8. doi: 10.1074/jbc.M104068200

    2001 doi

  16. [24]

    Gener- ative artificial intelligence in drug discovery: basic framework, recent advances, challenges, and opportunities.Front Pharmacol

    Gangwal A, Ansari A, Ahmad I, Azad AK, Kumarasamy V, Subramaniyan V, et al. Gener- ative artificial intelligence in drug discovery: basic framework, recent advances, challenges, and opportunities.Front Pharmacol. 2024;15:1331062. doi: 10.3389/fphar.2024.1331062

    2024 doi

  17. [25]

    Methods for the elucidation of protein–small molecule interactions.Chem Biol

    McFedries A, Schwaid A, Saghatelian A. Methods for the elucidation of protein–small molecule interactions.Chem Biol. 2013;20(5):667–73. doi: 10.1016/j.chembiol.2013.04.008

    2013 doi

  18. [26]

    Small-Molecule Target Engagement in Cells.Cell Chem Biol

    Schürmann M, Janning P, Ziegler S, Waldmann H. Small-Molecule Target Engagement in Cells.Cell Chem Biol. 2016;23(4):435–41. doi: 10.1016/j.chembiol.2016.03.008

    2016 doi

  19. [27]

    Transformer-based deep learning for predicting protein properties in the life sciences

    Chandra A, Tünnermann L, Löfstedt T, Gratz R. Transformer-based deep learning for predicting protein properties in the life sciences. Dötsch V, editor.eLife. 2023;12:e82819. doi: 10.7554/eLife.82819

    2023 doi

  20. [28]

    ADMET-score – a comprehensive scoring function for evaluation of chemical drug-likeness.MedChemComm

    Guan L, Yang H, Cai Y, Sun L, Di P, Li W, et al. ADMET-score – a comprehensive scoring function for evaluation of chemical drug-likeness.MedChemComm. 2019;10(1):148–

    2019

  21. [29]

    (Electronic supplementary information available.) doi: 10.1039/c8md00472b

    doi

  22. [30]

    Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions.J Cheminformatics

    Ertl P, Schuffenhauer A. Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions.J Cheminformatics. 2009;1:8. doi: 10.1186/1758-2946-1-8 14

    2009 doi

  23. [31]

    DrugBank: a comprehensive resource for in silico drug discovery and exploration.Nucleic Acids Res

    Wishart DS, Knox C, Guo AC, Shrivastava S, Hassanali M, Stothard P, et al. DrugBank: a comprehensive resource for in silico drug discovery and exploration.Nucleic Acids Res. 2006;34(Database issue):D668–72. doi: 10.1093/nar/gkj067

    2006 doi

  24. [32]

    CHEMBL database release 33

    Adasme Mora MF, Arcila Toro R, Blackshaw J, Bosc N, Corbett S, De Veij M, et al. CHEMBL database release 33. ChEMBL Database. 2011. doi: 10.6019/CHEMBL.database.33

    2011 doi

  25. [33]

    ZINC20 – A Free Ultralarge-Scale Chemical Database for Ligand Discovery.J Chem Inf Model

    Irwin JJ, Tang KG, Young J, Dandarchuluun C, Wong BR, Khurelbaatar M, et al. ZINC20 – A Free Ultralarge-Scale Chemical Database for Ligand Discovery.J Chem Inf Model. 2020;60(12):6065–73. doi: 10.1021/acs.jcim.0c00675

    2020 doi

  26. [34]

    UniProt: the Universal Protein Knowledgebase in 2023.Nucleic Acids Res

    UniProt Consortium. UniProt: the Universal Protein Knowledgebase in 2023.Nucleic Acids Res. 2023;51(D1):D523–31. doi: 10.1093/nar/gkac1052

    2023 doi

  27. [35]

    Basic local alignment search tool

    Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ. Basic local alignment search tool. J Mol Biol. 1990;215(3):403–10. doi: 10.1016/S0022-2836(05)80360-2

    1990 doi

  28. [36]

    huggingface/trl

    Hugging Face. huggingface/trl. Hugging Face; 2025. [accessed 29 Sept 2025] Available from: https://github.com/huggingface/trl

    2025

  29. [37]

    Release 2023_09_5 (Q3 2023) Release·rdkit/rdkit

    RDKit. Release 2023_09_5 (Q3 2023) Release·rdkit/rdkit. GitHub. [accessed 17 Aug 2025] Available from:https://github.com/rdkit/rdkit/releases/tag/Release_2023_09_5

    2023

  30. [38]

    Protein-Ligand Blind Docking Using QuickVina-W With Inter-Process Spatio-Temporal Integration.Sci Rep

    Hassan NM, Alhossary AA, Mu Y, Kwoh C-K. Protein-Ligand Blind Docking Using QuickVina-W With Inter-Process Spatio-Temporal Integration.Sci Rep. 2017;7(1):15451. doi: 10.1038/s41598-017-15571-7

    2017 doi

  31. [39]

    Protein and ligand prepa- ration: parameters, protocols, and influence on virtual screening enrichments.J Comput Aided Mol Des

    Sastry GM, Adzhigirey M, Day T, Annabhimoju R, Sherman W. Protein and ligand prepa- ration: parameters, protocols, and influence on virtual screening enrichments.J Comput Aided Mol Des. 2013;27(3):221–34. doi: 10.1007/s10822-013-9644-8

    2013 doi

  32. [40]

    Schrödinger Release 2025-3: LigPrep, Schrödinger, LLC, New York, NY, 2025

    2025

  33. [41]

    OPLS4: Improving Force Field Accuracy on Challenging Regimes of Chemical Space.J Chem Theory Com- put

    Lu C, Wu C, Ghoreishi D, Chen W, Wang L, Damm W, et al. OPLS4: Improving Force Field Accuracy on Challenging Regimes of Chemical Space.J Chem Theory Com- put. 2021;17(7):4291–300. doi: 10.1021/acs.jctc.1c00302

    2021 doi

  34. [42]

    Extra precision glide: docking and scoring incorporating a model of hydrophobic enclosure for protein–ligand complexes.J Med Chem

    Friesner RA, Murphy RB, Repasky MP, Frye LL, Greenwood JR, Halgren TA, et al. Extra precision glide: docking and scoring incorporating a model of hydrophobic enclosure for protein–ligand complexes.J Med Chem. 2006;49(21):6177–96. doi: 10.1021/jm051256o 15

    2006 doi

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.