Pith. sign in

REVIEW 4 major objections 4 minor 122 references

Large Language Models Transform Organic Synthesis From Reaction Prediction to Automation

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that large language models have become the orchestrating layer of organic synthesis — predicting reactions, planning routes, and directing robots — and that a systematic review of 2023–2025 work can map this transition and

desk verdict Not a usable survey: the claimed benchmarking is absent, and the text contains impossible numbers and fabricated-looking citations. read the letter →

arxiv 2508.05427 v1 pith:BABAJYCS submitted 2025-08-07 cs.AI

Kartar Kumar Lohana Tharwani , Rajesh Kumar , Sumita , Numan Ahmed , Yong Tang This is my paper
classification cs.AI
keywords largelanguagemodelsorganicsynthesisreactionpredictionretrosynthesisautonomousroboticplatformsChemLLMsafetybenchmarks
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

Large language models are no longer just reaction predictors; the paper argues they now function as the coordinating layer of organic synthesis, translating molecular strings into synthetic plans and even into instructions for robotic lab hardware. To back this, the paper presents what it calls the first systematic review of LLMs as synthesis orchestrators, covering advances from 2023 to 2025, with a task-oriented taxonomy and an evaluation of 15 models across 8 public datasets such as USPTO-50K, USPTO-MIT, and ChemBench. The paper's central message is that the bottleneck has moved from raw prediction accuracy to integration: hybrid LLM–graph-neural-network–reinforcement-learning systems, real-time spectral feedback, and safety gates are what turn a text generator into a dependable lab partner. It also documents persistent failure modes — condition blindness, chirality errors, data leakage, and dual-use hazards — and proposes a roadmap of open benchmarks, federated learning, and explainable interfaces. If the survey's map is right, chemists get a consolidated picture of which tools work and where automation can be trusted.

What carries the argument

The load-bearing object is the chemistry-tuned transformer, an LLM fine-tuned on reaction corpora so that it generates products or precursors as token sequences (SMILES or SELFIES). Around it the survey assembles a two-layer machinery: hybrid architectures that add a knowledge-graph reranker (SynAsk) or graph-neural-network validator (GNNs) to correct stereochemistry and condition awareness, and closed-loop robotic platforms (Coscientist, ChemCrow, LLM-RDF) that convert the model's textual protocols into executed experiments with inline sensor feedback. What this machinery does is shift the central task from 'predict the product' to 'plan, validate, and execute a synthesis,' which is the pap

What would settle it

Trace the benchmark table's headline numbers (ChemLLM Top-5 = 92.3% on ChemBench, SynAsk Top-1 = 87.1% on a 10k Reaxys subset, and the 102% value in Table X) to their cited sources and reproduce them on the public datasets; if any cannot be found or reproduced, the survey's central performance map is not reliable.

Watch

Extended reading notes

Core claim

The paper's core claim: chemistry-tuned transformer LLMs, which treat SMILES and SELFIES as token sequences, have moved from speculative aids to state-of-the-art tools for reaction prediction and retrosynthesis, and now form the cognitive core of autonomous synthesis platforms. It reports ChemLLM's Top-5 = 92.3% on ChemBench and SynAsk's Top-1 = 87.1% on a 10,000-reaction Reaxys subset, while noting a general ceiling near 65% Top-1 on USPTO-50K. It also describes closed-loop systems (Coscientist, ChemCrow) coupling LLM planning with robotic execution and inline analytics, claims order-of-magnitude faster optimization, and identifies safety filters and green-chemistry metrics as emerging desi

Load-bearing premise

The survey's map of the field is only as trustworthy as the reported model scores and the existence of the systems it describes, and the manuscript itself contains signs that undercut that trust — a 102% accuracy figure, a placeholder arXiv identifier, and case studies citing unrelated papers.

Editorial extensions

If this is right

  • If the survey is correct, LLM performance on standard reaction benchmarks has plateaued near 65% Top-1 on USPTO-50K, so further gains will come from multimodal conditioning (spectra, 3D conformers) and uncertainty quantification rather than from simply adding more reaction data.
  • Adoption of hybrid LLM + GNN + RL pipelines, rather than pure LLMs, becomes the evidence-based route to fixing chirality errors and condition blindness, two failure modes the paper quantifies at 15–20% error rates.
  • Autonomous synthesis platforms will increasingly be judged by their closed-loop integration — real-time HPLC/spectroscopy feedback and hardware interoperability — as much as by their route-planning accuracy.
  • Safety filters such as ChemSafetyBench-style dual-use screening and toxicity checks will become a prerequisite for deployment, since the paper reports a 12% risk that unguarded LLMs propose regulated or hazardous syntheses.
  • Open benchmarks and federated learning, not proprietary datasets alone, are the paper's proposed mechanism for closing the accessibility gap that currently leaves most academic labs unable to train or run large chemistry LLMs.

Reading between the lines

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

  • If the surveyed trend holds, the next competitive arena in synthesis automation is laboratory middleware: standardised APIs (like the described SynthLink) and digital-twin validation, because prediction models have become commoditised while robotic interoperability lags at 25–35% latency.
  • A testable extension: build a 'lab-readiness' scorecard that penalises condition blindness and chirality-invalid SMILES alongside Top-k accuracy; this would directly arbitrate which of the surveyed models deserve custody of a robot.
  • A reader should treat the benchmark table as directional: the manuscript itself reports a 102% Top-5 accuracy, cites a placeholder arXiv ID (2403.XXXXX), and attaches case studies (e.g., SynAuto) to unrelated references, so the numbers await verification.
  • The paper's emphasis on federated learning suggests a concrete industrial experiment: train a synthesis LLM across three geographically separated labs without sharing raw reaction data, and measure whether performance on rare reaction classes approaches that of a centrally trained model.
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

4 major / 4 minor

Summary. The manuscript is a survey of large language models applied to organic synthesis, claiming to be 'the first systematic review of LLMs as orchestrators of synthetic chemistry' for 2023–2025. It proposes a taxonomy of LLM applications, a 'benchmarking and validation' contribution evaluating 15 models on 8 public datasets, and a roadmap for responsible innovation. The text covers reaction prediction and retrosynthesis, synthesis planning, autonomous robotic platforms, sustainability, education, and safety. The paper's central value proposition is that it provides a trustworthy, comprehensive map of the field with validated performance numbers and reproducible comparisons.

Significance. If the factual content were reliable, the survey would fill a genuine need: it assembles many systems, tables, and case studies in one place, and its taxonomy could help orient new researchers. However, the empirical core of the paper is not reliable. The claimed benchmark evaluation is not actually performed in the manuscript: no evaluation protocol, code, or artifact is provided, the reported numbers are internally inconsistent, and at least one reported accuracy exceeds 100%. The citation record is also systematically unreliable, with a placeholder arXiv ID and multiple case studies citing unrelated papers. Because the paper's stated central contributions include 'Benchmarking and Validation,' these defects are load-bearing. The qualitative survey sections have some descriptive value, but they cannot compensate for the unverifiable quantitative claims. No machine-checked proofs, reproducible code, or independently checkable derivations are supplied, so the contribution as submitted does not meet the evidentiary standard for a trustworthy map of the field.

major comments (4)
  1. [§VI.4 (Table X)] The passage on 'Equity and Accessibility' states that the open-source ChemLLM Community Edition 'trails by 12% in Top-5 accuracy (90% vs. 102% for commercial models).' An accuracy above 100% is impossible. This is not a trivial typo: the same passage presents these numbers as quantitative evidence in a 'data-driven evaluation,' and Table X repeats similar quantitative challenge metrics. A reader cannot trust the central benchmark narrative when a headline comparison contains an impossible value.
  2. [Tables IV and VI; §III.D] The performance tables are internally inconsistent. Table VI reports SynAsk ChemBench Top-1 = 82.6%, while Table IV and §III.D report SynAsk Top-1 = 87.1% on a Reaxys subset; Table IV lists ChemLLM Top-1 = 68.5% identically for both ChemBench and USPTO-MIT, which is implausible without explanation. The manuscript gives no split definitions, decoding settings, evaluation code, or per-dataset protocols. The abstract's claim to 'evaluate 15 state-of-the-art models on 8 public datasets' is therefore unsupported: the paper restates or presents numbers without the methodology needed to reproduce or verify them.
  3. [References [46], [96], [17], [114]] The reference list contains a placeholder arXiv ID: [46] is 'arXiv:2403.XXXXX'. More seriously, several case studies cite unrelated papers. For example, SynAuto (Pfizer, 2025) in §V.B.1 is supported by [96], which is the Paxlovid manufacturing paper, not a description of SynAuto; GreenRoute in §IV.A.4 cites [17], the ChemCrow preprint; and 'ChemLLM Community Edition' in §VI.4 and §VII.D cites [114], a paper about GEB-1.3B. Since these citations are the only evidence that the named systems exist and perform as claimed, the survey's map cannot be independently checked.
  4. [§V.A, §VI, and Introduction] The paper promises comparisons of 'accuracy, computational cost, and experimental reproducibility' across 15 models, but the body does not contain such a systematic comparison. Section V is largely narrative case studies, and Section VI aggregates heterogeneous numbers without explaining how they were collected, normalized, or verified. This mismatch between the stated contribution and the actual content is substantial: the claimed benchmarking/validation contribution is one of the paper's three headline contributions and is not delivered in a checkable form.
minor comments (4)
  1. [Notation and structure] Section numbering is inconsistent: Roman-numeral top-level sections (VI, VII) are interleaved with decimal subsections (6.1, 6.4). This makes cross-referencing difficult.
  2. [Figures] Several figure references are invalid, e.g., 'Fig. IV-C' is referenced but no such figure exists, and 'showing if Figure 8' appears to be an incomplete phrase. The full text also contains an unresolved inserted passage ('Thought for 7 seconds' followed by LaTeX tokens), which suggests the manuscript was not carefully prepared.
  3. [Table III] Table III (molecular-image datasets) is not discussed in the body text and has a typo in the caption ('grouped bysynthetic'). Either integrate the table into the narrative or remove it.
  4. [Reference list] The reference list contains duplicates and near-duplicates, e.g., [23] and [31] are both the GPT-4 Technical Report, and several SynAsk entries appear with different reference numbers. A consolidated, verified reference list is needed.

Circularity Check

1 steps flagged · score 6.0 of 10

Benchmarking contribution restates the surveyed papers' own metrics; no independent evaluation is performed.

  1. renaming known result [Section I (contribution list); Section III.D; Table IV; Section III.G]
    "Benchmarking and Validation: We evaluate 15 state-of-the-art models on 8 public datasets (e.g., USPTO, CASP [14]), comparing accuracy, computational cost, and experimental reproducibility. ... Recent data indicate a Top-1 accuracy improvement of 5.2% over ChemLLM, reaching 87.1% on a subset of 10,000 curated reactions from Reaxys. ... Table IV: Updated Performance Metrics for Reaction Prediction Models — SynAsk Reaxys (10k) 87.1 95.0 0.90. ... Table V summarizes ... typical state-of-the-art (SOTA) values for these metrics based on recent literature."

    The claimed benchmarking contribution is not an independent evaluation: no evaluation protocol, dataset split definitions, decoding settings, or code are provided, and the paper's own tables reproduce the same numbers that the text already attributes to the surveyed model papers. For example, ChemLLM's Top-5=92.3% on ChemBench is quoted from [57] in Sections II.B/III.D and then appears again as an 'updated' metric in Table IV; SynAsk's Reaxys Top-1=87.1% is the same value quoted from [43]. The survey's 'prediction' that LLMs perform well is therefore the surveyed papers' self-reported numbers, relabeled as a benchmark comparison rather than derived or independently verified.

full rationale

The only substantive circularity is in the benchmarking/validation contribution: the survey's performance tables are constructed by relabeling numbers taken from the model papers themselves (ChemLLM [57], SynAsk [43], etc.), with no evaluation protocol, code, or independent computation. This is a 'renaming known result' pattern: literature-reported metrics are renamed as the survey's own 'Updated Performance Metrics.' The paper's other components—taxonomy, background, roadmap—are qualitative and not derived from those numbers, so the circularity is partial, not total. I did not score higher (8-10) because there is no self-citation chain or definitional equivalence; the central map still has independent organizational content. Separately, the paper has serious correctness problems (impossible 102% accuracy in §6.4; conflicting Top-1 values between Table IV and Table VI; placeholder reference [46] arXiv:2403.XXXXX), but those are integrity/verification risks, not circularity, and are noted only to explain why the restated benchmarks cannot be treated as validated.

Assumptions & free parameters 0 free parameters · 3 assumptions · 4 invented entities

The central claim of a trustworthy survey rests on the accuracy of the cited literature and the existence of the described systems. Both premises are violated by the smoking-gun evidence of placeholder arXiv IDs, impossible percentages, and mismatched case-study citations. Free parameters: none, since no new fit is performed.

assumptions (3)
  • domain assumption Reported performance metrics in the cited papers are accurate and correctly attributed
    The survey's summaries and Table IV rely on these numbers; several are unverifiable or mismatched (e.g., 102% accuracy in Section VI.4).
  • domain assumption The described systems and case studies exist as described
    Examples such as SynAuto, ChemBot (2026), and Guardian-LLM are presented as real; the cited references do not support them.
  • ad hoc to paper A benchmark evaluation of 15 models on 8 datasets was performed
    The Introduction claims a benchmarking contribution, but no protocol, results, or artifacts are presented.
invented entities (4)
  • SynAuto (Pfizer, 2025)
    purpose: Example of end-to-end autonomous synthesis
    Cited ref [96] is about Paxlovid manufacturing, not SynAuto; no other evidence.
  • ChemBot (2026)
    purpose: Example of LLM-native robotic chemistry
    Cited ref [106] is Organa (2024); future date 2026 is impossible for this 2025 preprint.
  • GreenRoute
    purpose: Example of green chemistry optimization
    No citation provided; cannot be verified.
  • Guardian-LLM
    purpose: Compliance filter for hazardous synthesis
    No verifiable citation; presented as a product without source.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Large Language Models Transform Organic Synthesis From Reaction Prediction to Automation." pith.science (2026). https://pith.science/paper/BABAJYCS

@misc{pith2026250805427,
  author       = {Pith},
  title        = {Pith review of: Large Language Models Transform Organic Synthesis From Reaction Prediction to Automation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BABAJYCS}},
  note         = {Machine review of arXiv:2508.05427}
}
read the original abstract

Large language models (LLMs) are beginning to reshape how chemists plan and run reactions in organic synthesis. Trained on millions of reported transformations, these text-based models can propose synthetic routes, forecast reaction outcomes and even instruct robots that execute experiments without human supervision. Here we survey the milestones that turned LLMs from speculative tools into practical lab partners. We show how coupling LLMs with graph neural networks, quantum calculations and real-time spectroscopy shrinks discovery cycles and supports greener, data-driven chemistry. We discuss limitations, including biased datasets, opaque reasoning and the need for safety gates that prevent unintentional hazards. Finally, we outline community initiatives open benchmarks, federated learning and explainable interfaces that aim to democratize access while keeping humans firmly in control. These advances chart a path towards rapid, reliable and inclusive molecular innovation powered by artificial intelligence and automation.

Figures

Figures reproduced from arXiv: 2508.05427 by the authors.

Figure 1
Figure 1. Taxonomy of current approaches for transferring general LLMs to specialized chemical LLMs. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Key challenges in applying LLMs to organic synthesis, including data limitations and ethical [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of LLM transformer architecture, highlighting attention mechanisms. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Rule-based system vs. LLMs for the prediction [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Comparative performance of LLMs on chem [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Reaction prediction using ChemLLM. Input SMILES strings for ethanol ( [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Overview of Chemical Sequence Modeling Approaches [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Autonomous Robotic Platform [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: SMILES to Product Flow [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: LLM-guided case studies from ChemCrow, SynAuto, and RoboRXN. [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: Ethical Safeguard Pipeline Reinforcement learning (RL) frameworks incorpo￾rating experimental feedback could refine synthetic pathway optimization, while integration with auto￾mated laboratory platforms may enable real-world performance validation. Such hybrid systems…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

122 extracted references · 63 canonical work pages

  1. [13]

    A review of large language models and autonomous agents in chem- istry

    M. C. Ramos, C. J. Collison, and A. D. White, “A review of large language models and autonomous agents in chem- istry.” arXiv preprint, 2024. arXiv:2407.01603

    arXiv 2024

  2. [46]

    Bidirectional chemical intelligent net: A uni- fied deep learning-based framework for predicting chemi- cal reactions,

    X. Wang et al., “Bidirectional chemical intelligent net: A uni- fied deep learning-based framework for predicting chemi- cal reactions,” arXiv preprint, 2024. arXiv:2403.XXXXX

    2024

  3. [96]

    The unprecedented paxlovid journey from mil- ligrams to millions of patient doses during the covid-19 pandemic,

    W. Yu, M. K. Krishnan, M. Weekly, R. M. Shanker , P. Doshi, J. A. Ragan, R. A. Greene, B. Gampper , S. Caron, A. McKil- lop, et al., “The unprecedented paxlovid journey from mil- ligrams to millions of patient doses during the covid-19 pandemic,” Communications Medicine , vol. 5, no. 1, p. 80, 2025

    work page 2025

  4. [17]

    ChemCrow: Augmenting Large-Language Models with Chemistry Tools,

    A. M. Bran et al., “ChemCrow: Augmenting Large-Language Models with Chemistry Tools,” 2023. Preprint

    2023

  5. [3]

    Planning chemical syntheses with deep neural networks and symbolic ai,

    M. H. Segler , M. Preuss, and M. P. Waller , “Planning chemical syntheses with deep neural networks and symbolic ai,” Nature, vol. 555, no. 7698, pp. 604–610, 2018

    2018

  6. [4]

    Attention is all you need,

    A. Vaswani, N. Shazeer , N. Parmar , J. Uszkoreit, L. Jones, A. N. Gomez, Łukasz Kaiser , and I. Polosukhin, “Attention is all you need,” in Advances in Neural Information Processing Systems, vol. 30, 2017

    2017

  7. [6]

    Augment- ing large language models with chemistry tools,

    A. M. Bran, S. Cox, A. D. White, and P. Schwaller , “Augment- ing large language models with chemistry tools,” Nature Machine Intelligence, 2023

    2023

  8. [7]

    Au- tonomous Chemical Research with Large Language Mod- els,

    D. A. Boiko, R. MacKnight, B. Kline, and G. Gomes, “Au- tonomous Chemical Research with Large Language Mod- els,” Nature, vol. 624, no. 7992, pp. 570–578, 2023

    2023

Show all 122 references
  1. [8]

    Self-Referencing Embedded Strings (SELF- IES): A 100% Robust Molecular String Representation,

    M. Krenn et al., “Self-Referencing Embedded Strings (SELF- IES): A 100% Robust Molecular String Representation,” Ma- chine Learning: Science and Technology , 2020

    2020

  2. [9]

    Bert: Pre-training of deep bidirectional trans- formers for language understanding

    J. Devlin, “Bert: Pre-training of deep bidirectional trans- formers for language understanding.” arXiv preprint, 2018. arXiv:1810.04805

    2018 arXiv

  3. [10]

    Scientific large language models: A survey on biological & chemical domains

    Q. Zhang, K. Ding, T. Lyv, X. Wang, Q. Yin, Y. Zhang, J. Yu, Y. Wang, X. Li, Z. Xiang, and et al., “Scientific large language models: A survey on biological & chemical domains.” arXiv preprint, 2024. arXiv:2401.14656

    2024 arXiv

  4. [11]

    Unraveling Molecular Structure: A Multimodal Spectro- scopic Dataset for Chemistry,

    M. Alberts, O. Schilter , F. Zipoli, N. Hartrampf, and T. Laino, “Unraveling Molecular Structure: A Multimodal Spectro- scopic Dataset for Chemistry,” 2024. arXiv preprint arXiv:2407.17492

    2024 arXiv

  5. [12]

    Schnet: A Continuous-Filter Convolutional Neural Network for Modeling Quantum Interactions,

    K. Schütt et al., “Schnet: A Continuous-Filter Convolutional Neural Network for Modeling Quantum Interactions,” inAd- vances in Neural Information Processing Systems (NeurIPS) , 2017

    2017

  6. [14]

    What’s What: The (Nearly) Definitive Guide to Reaction Role Assignment,

    N. Schneider et al. , “What’s What: The (Nearly) Definitive Guide to Reaction Role Assignment,” Journal of Chemical Information and Modeling , 2016

    2016

  7. [15]

    Exploring the limits of transfer learning with a unified text-to-text transformer ,

    C. Raffel, N. Shazeer , A. Roberts, K. Lee, S. Narang, M. Matena, Y. Zhou, W. Li, and P. J. Liu, “Exploring the limits of transfer learning with a unified text-to-text transformer ,” Journal of Machine Learning Research, vol. 21, no. 140, pp. 1– 67, 2020

    2020

  8. [16]

    Chemllm: A chemical large language model,

    B. Zhang, Y. Guo, S. Xin, C. Chen, X. Hong, Y. Wang, and B. Li, “Chemllm: A chemical large language model,” 2024

    2024

  9. [18]

    “found in translation

    P. Schwaller , T. Gaudin, D. Lanyi, C. Bekas, and T. Laino, ““found in translation”: predicting outcomes of complex organic chemistry reactions using neural sequence-to- sequence models,” Chemical science, vol. 9, no. 28, pp. 6091– 6098, 2018

    2018

  10. [19]

    An automatic end-to-end chemical synthesis development platform powered by large language models,

    S. Li, J. Hu, X. Gao, Y. Guo, Z. Yang, J. Li, L. Wang, and W. Zhang, “An automatic end-to-end chemical synthesis development platform powered by large language models,” Nature Communications, 2024

    2024

  11. [21]

    Chatmof: a neurosymbolic ai approach to accelerate metal-organic framework research with large language models,

    Y. Lee, J. Kim, S. Kim, J. Kim, and S. Choi, “Chatmof: a neurosymbolic ai approach to accelerate metal-organic framework research with large language models,” Chemical Science, 2025

    2025

  12. [22]

    Leveraging large language models for predic- tive chemistry,

    K. M. Jablonka, P. Schwaller , A. Ortega-Guerrero, and B. Smit, “Leveraging large language models for predic- tive chemistry,” Nature Machine Intelligence , vol. 6, no. 2, pp. 161–169, 2024

    2024

  13. [23]

    GPT-4 Technical Report,

    J. Achiam, S. Adler , S. Agarwal, L. Ahmad, et al. , “GPT-4 Technical Report,” 2023. arXiv preprint arXiv:2303.08774

    2023 arXiv

  14. [24]

    Do Large Lan- guage Models Understand Chemistry? A Conversation with ChatGPT,

    C. M. Castro Nascimento and A. S. Pimentel, “Do Large Lan- guage Models Understand Chemistry? A Conversation with ChatGPT,” Journal of Chemical Information and Modeling , vol. 63, no. 6, pp. 1649–1655, 2023

    2023

  15. [25]

    Large language models for chemistry robotics,

    N. Yoshikawa, M. Skreta, K. Darvish, S. Arellano-Rubach, Z. Ji, L. B. Kristensen, A. Z. Li, Y. Zhao, H. Xu, A. Kuramshin, and et al., “Large language models for chemistry robotics,” Autonomous Robots, vol. 47, no. 8, pp. 1057–1086, 2023

    2023

  16. [26]

    PubChem Substance and Compound Databases,

    S. Kim et al. , “PubChem Substance and Compound Databases,” Nucleic Acids Research , 2016

    2016

  17. [27]

    Moleculenet: A benchmark for molecular machine learning,

    Z. Wu, B. Ramsundar , E. N. Feinberg, J. Gomes, C. Geniesse, A. S. Pappu, K. Leswing, and V. Pande, “Moleculenet: A benchmark for molecular machine learning,” Chemical Sci- ence, vol. 9, no. 2, pp. 513–530, 2018

    2018

  18. [28]

    Transfer learning enables the molecular transformer to predict regio- and stereoselective reactions on carbohy- drates,

    G. Pesciullesi, P. Schwaller , T. Laino, and J.-L. Reymond, “Transfer learning enables the molecular transformer to predict regio- and stereoselective reactions on carbohy- drates,” Nature Communications, vol. 11, no. 1, p. 4874, 2020

    2020

  19. [29]

    Chemllm: A chemical large language model,

    D. Zhang, J. Zhang, H. Jiang, H. Yang, and Y. Yao, “Chemllm: A chemical large language model,” arXiv preprint, 2024

    2024

  20. [30]

    Chemformer: A pre-trained transformer for computational chemistry,

    R. Irwin, S. Dimitriadis, J. He, E. J. Bjerrum, and S. Genheden, “Chemformer: A pre-trained transformer for computational chemistry,” Chemical Science, 2022

    2022

  21. [31]

    GPT-4 technical report,

    J. Achiam, S. Adler , S. Agarwal, L. Ahmad, et al. , “GPT-4 technical report,” arXiv preprint, 2023

    2023

  22. [32]

    The rise of auto- mated curiosity-driven discoveries in chemistry,

    L. Bustillo, T. Laino, and T. Rodrigues, “The rise of auto- mated curiosity-driven discoveries in chemistry,” Chemical Science, vol. 14, pp. 10378–10384, 2023

    2023

  23. [33]

    Self-referencing embedded strings (SELFIES): A 100% robust molecular string representation,

    M. Krenn, F. Häse, A. Nigam, P. Friederich, and A. Aspuru- Guzik, “Self-referencing embedded strings (SELFIES): A 100% robust molecular string representation,” Machine Learning: Science and Technology , vol. 1, no. 4, p. 045024, 2020

    2020

  24. [34]

    Mol- instructions: A large-scale biomolecular instruction dataset for large language models,

    Y. Fang, Q. Zhang, H. Yang, X. Zhuang, et al. , “Mol- instructions: A large-scale biomolecular instruction dataset for large language models,” arXiv preprint, 2023

    2023

  25. [35]

    Smiles, a chemical language and information system. 1. introduction to methodology and encoding rules,

    D. Weininger , “Smiles, a chemical language and information system. 1. introduction to methodology and encoding rules,” Journal of Chemical Information and Computer Sciences , vol. 28, no. 1, pp. 31–36, 1988

    1988

  26. [36]

    Found in translation: Predicting outcomes of complex organic chemistry reactions using neural sequence-to- sequence models,

    P. Schwaller , T. Gaudin, D. Lanyi, C. Bekas, and T. Laino, “Found in translation: Predicting outcomes of complex organic chemistry reactions using neural sequence-to- sequence models,” Chemical Science , vol. 9, pp. 6091–6098, 2018

    2018

  27. [37]

    Hallucination is in- evitable: An innate limitation of large language models

    Z. Xu, S. Jain, and M. Kankanhalli, “Hallucination is in- evitable: An innate limitation of large language models.” arXiv preprint, 2024. arXiv:2401.11817

    2024 arXiv

  28. [38]

    Are large language models superhuman chemists?,

    A. D. White, G. M. Hocky, H. A. Gandhi, M. Ghaani, Y. Wang, and T. Yu, “Are large language models superhuman chemists?,” 2024

    2024

  29. [39]

    Chemberta: large-scale self-supervised pretraining for molecular prop- erty prediction,

    S. Chithrananda, G. Grand, and B. Ramsundar , “Chemberta: large-scale self-supervised pretraining for molecular prop- erty prediction,” arXiv preprint arXiv:2010.09885, 2020

    2010 arXiv

  30. [40]

    Mol-instructions: A large-scale biomolecu- lar instruction dataset for large language models

    Y. Fang, X. Liang, N. Zhang, K. Liu, R. Huang, Z. Chen, X. Fan, and H. Chen, “Mol-instructions: A large-scale biomolecu- lar instruction dataset for large language models.” arXiv preprint, 2023. arXiv:2306.08018

    2023 arXiv

  31. [41]

    Text2mol: Cross-modal molecule retrieval with natural language queries,

    C. Edwards, C. Zhai, and H. Ji, “Text2mol: Cross-modal molecule retrieval with natural language queries,” in Pro- ceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pp. 595–607, 2021. 21

    2021

  32. [42]

    MolCA: Molecular Graph-Language Modeling with Cross-Modal Projector and Uni-Modal Adapter,

    Z. Liu et al., “MolCA: Molecular Graph-Language Modeling with Cross-Modal Projector and Uni-Modal Adapter,” in Proceedings of the EMNLP , 2023

    2023

  33. [43]

    Synask: unleashing the power of large language models in organic synthesis,

    Z. Liu, Z. Wang, Y. Fang, Y. Gao, and L. Wang, “Synask: unleashing the power of large language models in organic synthesis,” Chemical Science, 2025

    2025

  34. [44]

    Large language models to accelerate organic chemistry synthesis,

    Z. Huang, T. Zhang, Y. Fang, and L. Wang, “Large language models to accelerate organic chemistry synthesis,” 2025

    2025

  35. [45]

    Attention is all you need,

    A. Vaswani et al. , “Attention is all you need,” Advances in Neural Information Processing Systems , vol. 30, 2017

    2017

  36. [47]

    Molecule edit graph attention network: Modeling chemical reactions as sequences of graph edits,

    M. Sacha et al. , “Molecule edit graph attention network: Modeling chemical reactions as sequences of graph edits,” Journal of Chemical Information and Modeling, vol. 61, no. 7, pp. 3273–3284, 2021

    2021

  37. [48]

    Chembench: A publicly accessible, integrated cheminformatics portal,

    S. J. Capuzzi et al. , “Chembench: A publicly accessible, integrated cheminformatics portal,” Journal of Chemical Information and Modeling , vol. 57, no. 2, pp. 105–108, 2017

    2017

  38. [49]

    Orderly: Data sets and benchmarks for chem- ical reaction data,

    D. S. Wigh, J. Arrowsmith, A. Pomberger , K. C. Felton, and A. A. Lapkin, “Orderly: Data sets and benchmarks for chem- ical reaction data,” Journal of Chemical Information and Modeling, vol. 64, no. 10, pp. 3790–3798, 2024

    2024

  39. [50]

    MolNexTR: A Generalized Deep Learning Model for Molecular Image Recognition,

    Y. Chen et al. , “MolNexTR: A Generalized Deep Learning Model for Molecular Image Recognition,” 2024. arXiv preprint arXiv:2403.03691

    2024 arXiv

  40. [51]

    Molgrapher: Graph-based visual recognition of chemical structures,

    L. Morin, M. Danelljan, M. I. Agea, A. Nassar , V. Weber , I. Meijer , P. Staar , and F. Yu, “Molgrapher: Graph-based visual recognition of chemical structures,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 19552–19561, 2023

    2023

  41. [52]

    Img2mol– accurate smiles recognition from molecular graphical de- pictions,

    D.-A. Clevert, T. Le, R. Winter , and F. Montanari, “Img2mol– accurate smiles recognition from molecular graphical de- pictions,” Chemical Science, vol. 12, no. 42, pp. 14174–14181, 2021

    2021

  42. [53]

    Chemvlm: Exploring the power of multimodal large language models in chemistry,

    J. Li, D. Zhang, X. Wang, Z. Hao, J. Lei, Q. Tan, C. Zhou, W. Liu, Y. Yang, X. Xiong, et al. , “Chemvlm: Exploring the power of multimodal large language models in chemistry,” Proceedings of the AAAI Conference on Artificial Intelligence , vol. 39, no. 1, pp. 415–423, 2025

    2025

  43. [54]

    Rxnscribe: A sequence generation model for reaction diagram pars- ing,

    Y. Qian, J. Guo, Z. Tu, C. W. Coley, and R. Barzilay, “Rxnscribe: A sequence generation model for reaction diagram pars- ing,” Journal of Chemical Information and Modeling , vol. 63, no. 13, pp. 4030–4041, 2023

    2023

  44. [55]

    Openchemie: An information extraction toolkit for chemistry literature,

    V. Fan, Y. Qian, A. Wang, A. Wang, C. W. Coley, and R. Barzi- lay, “Openchemie: An information extraction toolkit for chemistry literature,” Journal of Chemical Information and Modeling, 2024

    2024

  45. [56]

    Reactiondataextractor 2.0: A deep learning approach for data extraction from chemical reaction schemes,

    D. M. Wilary and J. M. Cole, “Reactiondataextractor 2.0: A deep learning approach for data extraction from chemical reaction schemes,” Journal of Chemical Information and Modeling, vol. 63, no. 19, pp. 6053–6067, 2023

    2023

  46. [57]

    Chemllm: A chemical large language model,

    D. Zhang et al. , “Chemllm: A chemical large language model,” arXiv preprint arXiv:2402.06852, 2024

    2024 arXiv

  47. [58]

    Chemformer: a pre-trained transformer for computational drug discovery,

    R. Irwin, S. Dimitriadis, J. He, E. J. Bjerrum, and S. Genheden, “Chemformer: a pre-trained transformer for computational drug discovery,” Chemical Science, 2025

    2025

  48. [59]

    Molecular transformer: a model for uncertainty-calibrated chemical reaction prediction,

    P. Schwaller , T. Laino, T. Gaudin, P. Bolgar , C. A. Hunter , C. Bekas, and A. A. Lee, “Molecular transformer: a model for uncertainty-calibrated chemical reaction prediction,” ACS central science, vol. 5, no. 9, pp. 1572–1583, 2019

    2019

  49. [60]

    Chemformer: A Pre-Trained Transformer for Computational Chemistry,

    R. Irwin et al., “Chemformer: A Pre-Trained Transformer for Computational Chemistry,” Machine Learning: Science and Technology, 2022

    2022

  50. [61]

    Using machine learning to predict suitable conditions for organic reactions,

    H. Gao, T. J. Struble, C. W. Coley, Y. Wang, W. H. Green, and K. F. Jensen, “Using machine learning to predict suitable conditions for organic reactions,” ACS central science, vol. 4, no. 11, pp. 1465–1476, 2018

    2018

  51. [62]

    Trans- fer learning: making retrosynthetic predictions based on a small chemical reaction dataset scale to a new level,

    R. Bai, C. Zhang, L. Wang, C. Yao, J. Ge, and H. Duan, “Trans- fer learning: making retrosynthetic predictions based on a small chemical reaction dataset scale to a new level,” Molecules, vol. 25, no. 10, p. 2357, 2020

    2020

  52. [63]

    Bidirectional chemical intelligent net: A unified deep learning–based framework for predicting chemical reaction,

    X. Xue, K. Chen, H. Sun, X. Liu, X. Liu, S. Li, J. Yan, Y. Peng, M. S. Mubarak, A. Al-Harrasi, et al., “Bidirectional chemical intelligent net: A unified deep learning–based framework for predicting chemical reaction,” Chinese Chemical Letters, p. 110968, 2025

    2025

  53. [64]

    Molecule edit graph attention network: mod- eling chemical reactions as sequences of graph edits,

    M. Sacha, M. Błaz, P. Byrski, P. Dabrowski-Tumanski, M. Chrominski, R. Loska, P. Włodarczyk-Pruszynski, and S. Jastrzebski, “Molecule edit graph attention network: mod- eling chemical reactions as sequences of graph edits,” Jour- nal of Chemical Information and Modeling , vol....

    2021

  54. [65]

    Are large language models superhuman chemists?,

    A. Mirza, N. Alampara, S. Kunchapu, M. Ríos-García, B. Emoekabu, A. Krishnan, T. Gupta, M. Schilling-Wilhelmi, M. Okereke, A. Aneesh, et al. , “Are large language models superhuman chemists?,” arXiv preprint arXiv:2404.01475 , 2024

    2024 arXiv

  55. [66]

    Enhancing chemical synthesis: a two-stage deep neural network for predicting feasible reac- tion conditions,

    L.-Y. Chen and Y.-P. Li, “Enhancing chemical synthesis: a two-stage deep neural network for predicting feasible reac- tion conditions,” Journal of Cheminformatics , vol. 16, no. 1, p. 11, 2024

    2024

  56. [67]

    Federated learning in chem- ical engineering: A tutorial on a framework for privacy- preserving collaboration across distributed data sources,

    S. Dutta, I. Leal de Freitas, P. Maciel Xavier , C. Miceli de Farias, and D. E. Bernal Neira, “Federated learning in chem- ical engineering: A tutorial on a framework for privacy- preserving collaboration across distributed data sources,” Industrial & Engineering Chemistry Res...

    2025

  57. [68]

    Hybrid-llm-gnn: integrating large language models and graph neural net- works for enhanced materials property prediction,

    Y. Li, V. Gupta, M. N. T. Kilic, K. Choudhary, D. Wines, W.- k. Liao, A. Choudhary, and A. Agrawal, “Hybrid-llm-gnn: integrating large language models and graph neural net- works for enhanced materials property prediction,” Digital Discovery, 2025

    2025

  58. [69]

    In silico toxicology: compu- tational methods for the prediction of chemical toxicity,

    A. B. Raies and V. B. Bajic, “In silico toxicology: compu- tational methods for the prediction of chemical toxicity,” Wiley Interdisciplinary Reviews: Computational Molecular Science, vol. 6, no. 2, pp. 147–172, 2016

    2016

  59. [70]

    Prompt sentiment: The catalyst for llm change,

    V. Gandhi and S. Gandhi, “Prompt sentiment: The catalyst for llm change,” arXiv preprint arXiv:2503.13510, 2025

    2025 arXiv

  60. [71]

    Automation and machine learning augmented by large language models in a catalysis study,

    Y. Su, X. Wang, Y. Ye, Y. Xie, Y. Xu, Y. Jiang, and C. Wang, “Automation and machine learning augmented by large language models in a catalysis study,” Chemical Science , vol. 15, no. 31, pp. 12200–12233, 2024

    2024

  61. [72]

    A chemist’s guide to multi-objective optimization solvers for reaction optimization,

    A. S. Vel, D. Cortés-Borda, and F.-X. Felpin, “A chemist’s guide to multi-objective optimization solvers for reaction optimization,” Reaction Chemistry & Engineering , vol. 9, no. 11, pp. 2882–2891, 2024

    2024

  62. [73]

    Finding environmental-friendly chemical synthesis with ai and high-throughput robotics,

    V.-H. Vu, K.-H. Bui, K. D. Dang, M. Duong-Tuan, D. D. Le, and T. Nguyen-Dang, “Finding environmental-friendly chemical synthesis with ai and high-throughput robotics,” Journal of Science: Advanced Materials and Devices , vol. 10, no. 1, p. 100818, 2025

    2025

  63. [74]

    The rise of auto- mated curiosity-driven discoveries in chemistry,

    L. Bustillo, T. Laino, and T. Rodrigues, “The rise of auto- mated curiosity-driven discoveries in chemistry,” Chemical Science, vol. 14, no. 38, pp. 10378–10384, 2023

    2023

  64. [75]

    Real-time adaptive synthesis op- timization using large language models,

    H. Kim, S. Lee, and J. Park, “Real-time adaptive synthesis op- timization using large language models,” Chemical Science, vol. 15, pp. 3456–3467, 2024

    2024

  65. [76]

    Recent advances in se- quentially pd-catalyzed one-pot syntheses of heterocycles,

    M. M. Kornet and T. J. Müller , “Recent advances in se- quentially pd-catalyzed one-pot syntheses of heterocycles,” Molecules, vol. 29, no. 22, p. 5265, 2024

    2024

  66. [77]

    Novel methodologies for chemical activation in organic synthesis under solvent-free reaction conditions,

    C. G. Avila-Ortiz and E. Juaristi, “Novel methodologies for chemical activation in organic synthesis under solvent-free reaction conditions,” Molecules, vol. 25, no. 16, p. 3579, 2020

    2020

  67. [78]

    Chemberta-2: Towards chemical foundation mod- els,

    W. Ahmad, E. Simon, S. Chithrananda, G. Grand, and B. Ram- sundar , “Chemberta-2: Towards chemical foundation mod- els,” arXiv preprint arXiv:2209.01712, 2022

    2022 arXiv

  68. [79]

    Geom, Energy-Annotated Molecular Con- formations for Property Prediction and Molecular Genera- tion,

    S. Axelrod et al., “Geom, Energy-Annotated Molecular Con- formations for Property Prediction and Molecular Genera- tion,” Scientific Data, 2022

    2022

  69. [80]

    Selformer: Molecular representation learning via selfies language models,

    A. Yüksel, E. Ulusoy, A. Ünlü, and T. Doğan, “Selformer: Molecular representation learning via selfies language models,” Machine Learning: Science and Technology , vol. 4, no. 2, p. 025035, 2023

    2023

  70. [81]

    Regression Transformer Enables Concurrent Sequence Regression and Generation for Molecular Lan- guage Modelling,

    J. Born et al., “Regression Transformer Enables Concurrent Sequence Regression and Generation for Molecular Lan- guage Modelling,” Nature Machine Intelligence , 2023

    2023

  71. [83]

    A Deep-Learning System Bridging Molecule Structure and Biomedical Text with Comprehension Com- parable to Human Professionals,

    Z. Zeng et al., “A Deep-Learning System Bridging Molecule Structure and Biomedical Text with Comprehension Com- parable to Human Professionals,” Nature Communications, 2022

    2022

  72. [84]

    Molformer: Motif-Based Transformer on 3D Heterogeneous Molecular Graphs,

    F. Wu et al. , “Molformer: Motif-Based Transformer on 3D Heterogeneous Molecular Graphs,” in Proceedings of the AAAI, 2023

    2023

  73. [85]

    Biot5: Enriching cross-modal integration in biology with chemical knowledge and natural language associations,

    Q. Pei, W. Zhang, J. Zhu, K. Wu, K. Gao, L. Wu, Y. Xia, and R. Yan, “Biot5: Enriching cross-modal integration in biology with chemical knowledge and natural language associations,” arXiv preprint arXiv:2310.07276, 2023

    2023 arXiv

  74. [86]

    Au- tonomous chemical research with large language models,

    D. A. Boiko, R. MacKnight, B. Kline, and G. Gomes, “Au- tonomous chemical research with large language models,” Nature, vol. 624, no. 7992, pp. 570–578, 2023

    2023

  75. [87]

    Druggpt: a gpt-based strategy for designing potential ligands targeting specific proteins,

    Y. Li, C. Gao, X. Song, X. Wang, Y. Xu, and S. Han, “Druggpt: a gpt-based strategy for designing potential ligands targeting specific proteins,” bioRxiv, pp. 2023–06, 2023

    2023

  76. [88]

    A molecular multimodal foundation model associating molecule graphs with natural language

    B. Su, D. Du, Z. Yang, Y. Zhou, J. Li, A. Rao, H. Sun, Z. Lu, and J.-R. Wen, “A molecular multimodal foundation model associating molecule graphs with natural language.” arXiv preprint, 2022. arXiv:2209.05481

    2022 arXiv

  77. [89]

    Multi-Modal Molecule Structure–Text Model for Text-Based Retrieval and Editing,

    S. Liu et al. , “Multi-Modal Molecule Structure–Text Model for Text-Based Retrieval and Editing,” Nature Machine In- telligence, 2023

    2023

  78. [90]

    Unimap: universal smiles-graph representation learning,

    S. Feng, L. Yang, Y. Huang, Y. Ni, W. Ma, and Y. Lan, “Unimap: universal smiles-graph representation learning,” arXiv preprint arXiv:2310.14216, 2023

    2023 arXiv

  79. [91]

    Featur- izations matter: A multiview contrastive learning approach to molecular pretraining,

    Y. Zhu, D. Chen, Y. Du, Y. Wang, Q. Liu, and S. Wu, “Featur- izations matter: A multiview contrastive learning approach to molecular pretraining,” in Proceedings of the ICML AI for Science Workshop, 2022

    2022

  80. [92]

    Graph neural networks for molecules,

    Y. Wang, Z. Li, and A. Barati Farimani, “Graph neural networks for molecules,” in Machine learning in molecular sciences, pp. 21–66, Springer , 2023

    2023

  81. [93]

    Reinforcement learning for improving chemical reaction performance,

    A. Hoque, M. Surve, S. Kalyanakrishnan, and R. B. Sunoj, “Reinforcement learning for improving chemical reaction performance,” Journal of the American Chemical Society , vol. 146, no. 41, pp. 28250–28267, 2024

    2024

  82. [94]

    Language models for predict- ing organic synthesis procedures,

    K. Huang and R. N. Zare, “Language models for predict- ing organic synthesis procedures,” MDPI Applied Sciences , vol. 14, no. 24, p. 11526, 2024

    2024

  83. [95]

    De novo drug design using reinforcement learning with multiple GPT agents,

    X. Hu, G. Liu, Y. Zhao, and H. Zhang, “De novo drug design using reinforcement learning with multiple GPT agents,” in Advances in Neural Information Processing Systems , vol. 36, 2024

    2024

  84. [97]

    Decomposition for enhancing attention: Improving LLM-based text-to-sql through work- flow paradigm,

    Y. Xie, X. Jin, T. Xie, M. Matrixmxlin, L. Chen, C. Yu, C. Lei, C. Zhuo, B. Hu, and Z. Li, “Decomposition for enhancing attention: Improving LLM-based text-to-sql through work- flow paradigm,” in Findings of the Association for Computa- tional Linguistics: ACL 2024, (Bangkok, ...

    2024

  85. [98]

    Federated learning of molecular properties with graph neural networks in a het- erogeneous setting,

    W. Zhu, J. Luo, and A. D. White, “Federated learning of molecular properties with graph neural networks in a het- erogeneous setting,” Patterns, vol. 3, no. 6, 2022

    2022

  86. [99]

    Optimizing human–ai col- laboration in chemistry: A case study on enhancing genera- tive ai responses through prompt engineering,

    D. V. Vidhani and M. Mariappan, “Optimizing human–ai col- laboration in chemistry: A case study on enhancing genera- tive ai responses through prompt engineering,” Chemistry, vol. 6, no. 4, pp. 723–737, 2024

    2024

  87. [100]

    Unleashing the potential of llms for quan- tum computing: A study in quantum architecture design,

    Z. Liang, J. Cheng, R. Yang, H. Ren, Z. Song, D. Wu, X. Qian, T. Li, and Y. Shi, “Unleashing the potential of llms for quan- tum computing: A study in quantum architecture design,” arXiv preprint arXiv:2307.08191, 2023

    2023 arXiv

  88. [101]

    Edge ai-enhanced traffic monitoring and anomaly detec- tion using multimodal large language models,

    R. Peruski, A. Saroj, W. Zhou, S. Djouadi, and C. Cao, “Edge ai-enhanced traffic monitoring and anomaly detec- tion using multimodal large language models,” in Interna- tional Conference on Transportation and Development 2025 , pp. 429–438, 2025

    2025

  89. [102]

    Autonomous execution of highly reac- tive chemical transformations in the schlenkputer ,

    N. L. Bell, F. Boser , A. Bubliauskas, D. R. Willcox, V. S. Luna, and L. Cronin, “Autonomous execution of highly reac- tive chemical transformations in the schlenkputer ,” Nature Chemical Engineering, vol. 1, no. 2, pp. 180–189, 2024

    2024

  90. [103]

    Open source, modular platforms, and the chal- lenge of fragmentation,

    C. S. Yoo, “Open source, modular platforms, and the chal- lenge of fragmentation,” Criterion J. on Innovation , vol. 1, p. 619, 2016

    2016

  91. [104]

    Studying llm performance on closed-and open-source data,

    T. Ahmed, C. Bird, P. Devanbu, and S. Chakraborty, “Studying llm performance on closed-and open-source data,” arXiv preprint arXiv:2402.15100, 2024

    2024 arXiv

  92. [105]

    Lab-on-robot: Un- manned mass spectrometry robot for direct sample analysis in hazardous and radioactive environments,

    X. Liu, X. Liu, B. Li, X. Zhang, and B. Hu, “Lab-on-robot: Un- manned mass spectrometry robot for direct sample analysis in hazardous and radioactive environments,” Analytical Chemistry, vol. 97, no. 17, pp. 9126–9130, 2025

    2025

  93. [106]

    Organa: A robotic assistant for automated chemistry ex- perimentation and characterization

    K. Darvish, M. Skreta, Y. Zhao, N. Yoshikawa, S. Som, M. Bog- danovic, Y. Cao, H. Hao, H. Xu, A. Aspuru-Guzik, and et al., “Organa: A robotic assistant for automated chemistry ex- perimentation and characterization.” arXiv preprint, 2024. arXiv:2401.06949

    2024 arXiv

  94. [107]

    Dual- use concerns in ai-driven drug discovery,

    F. Urbina, F. Lentzos, C. Invernizzi, and S. Ekins, “Dual- use concerns in ai-driven drug discovery,” Nature Machine Intelligence, vol. 4, no. 3, pp. 189–191, 2022

    2022

  95. [109]

    An autonomous laboratory for the accelerated synthesis of novel materials,

    N. J. Szymanski, B. Rendy, Y. Fei, R. E. Kumar , T. He, D. Mil- sted, M. J. McDermott, M. Gallant, E. D. Cubuk, A. Merchant, and et al., “An autonomous laboratory for the accelerated synthesis of novel materials,” Nature, vol. 624, no. 7990, pp. 86–91, 2023

    2023

  96. [110]

    Large language models to accelerate organic chemistry synthesis,

    Z. Huang, T. Zhang, Y. Fang, and L. Wang, “Large language models to accelerate organic chemistry synthesis,” arXiv preprint arXiv:2504.18340, 2025

    2025 arXiv

  97. [111]

    Synask: Unleashing the power of large language models in organic synthesis,

    Z. Liu, Z. Wang, Y. Fang, Y. Gao, and L. Wang, “Synask: Unleashing the power of large language models in organic synthesis,” Chemical Science, vol. 16, pp. 43–56, 2025

    2025

  98. [112]

    Hybrid-llm-gnn: Integrating large language models and graph neural networks for enhanced materials property prediction,

    Y. Li et al. , “Hybrid-llm-gnn: Integrating large language models and graph neural networks for enhanced materials property prediction,” Digital Discovery, 2025

    2025

  99. [113]

    Large language models for chemistry robotics,

    N. Yoshikawa et al., “Large language models for chemistry robotics,” Autonomous Robots, vol. 47, pp. 1057–1086, 2023

    2023

  100. [115]

    Graph neural networks for molecules,

    Y. Wang, Z. Li, and A. Barati Farimani, “Graph neural networks for molecules,” Machine learning in molecular sciences, pp. 21–66, 2023

    2023

  101. [116]

    Using machine learning to predict suitable conditions for organic reactions,

    H. Gao et al. , “Using machine learning to predict suitable conditions for organic reactions,” ACS Central Science, vol. 4, no. 11, pp. 1465–1476, 2018

    2018

  102. [117]

    An automatic end-to-end chemical synthesis development platform powered by large language models,

    S. Li et al. , “An automatic end-to-end chemical synthesis development platform powered by large language models,” Nature Communications, vol. 15, p. 10160, 2024

    2024

  103. [118]

    Dual- use concerns in ai-driven drug discovery,

    F. Urbina, F. Lentzos, C. Invernizzi, and S. Ekins, “Dual- use concerns in ai-driven drug discovery,” Nature Machine Intelligence, vol. 4, pp. 189–191, 2022

    2022

  104. [119]

    Finding environmental-friendly chemical synthesis with ai and high-throughput robotics,

    V.-H. Vu et al. , “Finding environmental-friendly chemical synthesis with ai and high-throughput robotics,” Journal of Science: Advanced Materials and Devices , vol. 10, p. 100818, 2025

    2025

  105. [120]

    Ecosynth sheds light on chemistry,

    K. Huvaere, “Ecosynth sheds light on chemistry,” Green Processing and Synthesis , vol. 1, no. 6, pp. 533–534, 2012

    2012

  106. [121]

    Valid-mol: a systematic framework for validated llm-assisted molecular design,

    H. Hudan Nuha et al., “Valid-mol: a systematic framework for validated llm-assisted molecular design,” arXiv e-prints, pp. arXiv–2506, 2025

    2025

  107. [122]

    Chemical reasoning in llms unlocks steerable synthesis planning and reaction mechanism elucidation,

    A. M. Bran, T. A. Neukomm, D. P. Armstrong, Z. Jončev, and P. Schwaller , “Chemical reasoning in llms unlocks steerable synthesis planning and reaction mechanism elucidation,” 2025

    2025

  108. [123]

    Chemsafetybench: Bench- marking llm safety on chemistry domain,

    H. Zhao, X. Tang, Z. Yang, X. Han, X. Feng, Y. Fan, S. Cheng, D. Jin, Y. Zhao, A. Cohan, et al. , “Chemsafetybench: Bench- marking llm safety on chemistry domain,” arXiv preprint arXiv:2411.16736, 2024

    2024 arXiv

  109. [124]

    Synask: unleashing the power of large language models in organic synthesis,

    C. Zhang, Q. Lin, B. Zhu, H. Yang, X. Lian, H. Deng, J. Zheng, and K. Liao, “Synask: unleashing the power of large language models in organic synthesis,” Chemical Sci- ence, vol. 16, no. 1, pp. 43–56, 2025. 23

    2025

  110. [125]

    Augmenting large language models with chemistry tools,

    A. M. Bran, S. Cox, O. Schilter , C. Baldassari, A. D. White, and P. Schwaller , “Augmenting large language models with chemistry tools,” Nature Machine Intelligence , vol. 6, no. 5, pp. 525–535, 2024

    2024

  111. [126]

    As- sessment of chemistry knowledge in large language models that generate code,

    A. D. White, G. M. Hocky, H. A. Gandhi, M. Ansari, S. Cox, G. P. Wellawatte, S. Sasmal, Z. Yang, K. Liu, Y. Singh, et al. , “As- sessment of chemistry knowledge in large language models that generate code,” Digital Discovery, vol. 2, no. 2, pp. 368– 376, 2023

    2023

  112. [127]

    Geb-1.3 b: Open lightweight large language model,

    J. Wu, Y. Zhu, L. Shen, and X. Lu, “Geb-1.3 b: Open lightweight large language model,” arXiv preprint arXiv:2406.09900, 2024

    2024 arXiv

  113. [128]

    An automatic end-to-end chemical synthesis development platform powered by large language models,

    Y. Ruan, C. Lu, N. Xu, Y. He, Y. Chen, J. Zhang, J. Xuan, J. Pan, Q. Fang, H. Gao, et al. , “An automatic end-to-end chemical synthesis development platform powered by large language models,” Nature communications , vol. 15, no. 1, p. 10160, 2024

    2024

  114. [129]

    Evaluation of nontargeted mass spectral data acquisition strategies for water analysis and toxicity-based feature prioritization by ms2tox,

    P. Peets, M. B. Rian, J. W. Martin, and A. Kruve, “Evaluation of nontargeted mass spectral data acquisition strategies for water analysis and toxicity-based feature prioritization by ms2tox,” Environmental Science & Technology , vol. 58, no. 39, pp. 17406–17418, 2024

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.