Project Ariadne: Prompt-Conditioned Route Generation for Synthesis Planning

Anton Morgunov; Victor S. Batista

arxiv: 2606.24184 · v1 · pith:ETRBAFECnew · submitted 2026-06-23 · 💻 cs.LG

Project Ariadne: Prompt-Conditioned Route Generation for Synthesis Planning

Anton Morgunov , Victor S. Batista This is my paper

Pith reviewed 2026-06-26 00:34 UTC · model grok-4.3

classification 💻 cs.LG

keywords retrosynthetic planningprompt-conditioned generationdecoder-only modelsynthesis route generationconstrained retrosynthesissequence generation

0 comments

The pith

A single decoder-only model generates synthesis routes that obey depth and starting-material constraints when those are added to the prompt.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper demonstrates that retrosynthetic planning with constraints can be reframed as completing a single prompt sequence that encodes the target molecule, the constraints, and the route. On the chosen benchmark family a 24-layer checkpoint improves solvability scores when the prompt includes depth or required-leaf fields. The same model runs faster than a bidirectional search planner on required-leaf tasks while remaining comparable to other direct-generation methods on standard targets. This removes the need to train separate models for each planning specification.

Core claim

Ariadne represents the target, optional constraints such as route depth or required starting materials, and the multistep route as one prompt-completion sequence inside a decoder-only transformer. A single 24-layer checkpoint trained this way follows the supplied constraints, raising Solv-0 by 13.7 points for depth prompts and 31.2 points for required-leaf prompts on the RetroCast/PaRoutes mkt-cnv-160 family. The model also exceeds DESP on required-leaf Top-10 and Solv-0 in 24 GPU-minutes versus 6.8 GPU-hours and matches DMS Explorer XL on standard reconstruction at roughly half the inference time.

What carries the argument

The unified prompt-completion sequence that places target, constraints, and route inside one decoder-only generation task.

If this is right

One trained checkpoint handles multiple constraint types instead of requiring separate models for each specification.
Inference time drops below that of bidirectional search planners while meeting or exceeding their constrained-task metrics.
Route-holdout reconstruction improves relative to prior direct-generation baselines.
New constraint types can be added by extending the prompt format without architectural changes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Prompt conditioning may allow rapid testing of new planning rules by simply altering the prompt text rather than retraining.
The approach could be combined with larger language models to handle more complex multi-objective constraints in a single pass.
Practical adoption would still require independent route validation tools beyond the current solvability metrics.

Load-bearing premise

The assumption that Solv-0 and Top-10 scores on the selected benchmark family measure the practical quality of routes without needing extra post-hoc checks.

What would settle it

Generate routes under explicit depth or leaf prompts and compare them to a set of experimentally verified routes that are known to satisfy the same constraints.

Figures

Figures reproduced from arXiv: 2606.24184 by Anton Morgunov, Victor S. Batista.

read the original abstract

Retrosynthetic planning seeks to connect a target molecule to commercially available starting materials through a multistep route. Classical planners construct such routes by iteratively applying single-step reaction models within a search procedure; constrained variants often require specialized algorithms or architectural changes. Direct route generation reframes retrosynthesis as sequence generation, but existing direct-generation methods still train separate models for different planning specifications. We introduce Ariadne, a decoder-only route generator that represents the target, optional constraints, and route in one prompt-completion sequence. On the RetroCast/PaRoutes mkt-cnv-160 benchmark family, one 24-layer checkpoint follows route-depth and required-starting-material prompts: adding the corresponding prompt fields raises Solv-0 by 13.7 points for depth constraints and 31.2 points for required-leaf constraints. Ariadne also improves over DESP, a bidirectional search planner, on required-leaf Top-10 and Solv-0 in 24 GPU-minutes versus 6.8 GPU-hours. On standard reconstruction, Ariadne is comparable to DMS Explorer XL at about half the reported inference time. Across additional target-only benchmarks, Ariadne's clearest gains are on route-holdout reconstruction, whereas AiZynthFinder MCTS remains stronger on several Solv-0 comparisons. These results extend sequence generation from specialist retrosynthesis models to prompt-conditioned structural route generation. We release the codebase and training scripts to support further work, but do not introduce Tier-1--3 route checkers; those remain the main bottleneck before models of this kind can become useful to experimental chemists.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Ariadne shows one decoder-only model can take depth and leaf prompts for retrosynthesis routes, but supplies no direct count of whether those prompts are obeyed.

read the letter

The central claim is that a single 24-layer decoder-only model can handle multiple constraint types in route generation by stuffing them into the prompt, and that this lifts Solv-0 by 13.7 points on depth prompts and 31.2 on required-leaf prompts versus the no-prompt baseline. That framing is new relative to the specialist models cited. The work also reports faster inference than DESP on the required-leaf task and comparable reconstruction numbers to DMS Explorer at roughly half the time. Releasing the code and training scripts is useful for anyone who wants to test the setup themselves.

The soft spot is the missing verification step. Solv-0 and Top-10 are downstream metrics; they can rise even when many outputs ignore the depth limit or omit the required starting material. The abstract itself notes the lack of Tier-1–3 checkers as the remaining bottleneck, yet the headline numbers rest on the assumption that the prompt fields alone enforce compliance. No separate statistic (fraction of routes whose depth matches the prompt, or that contain the required leaf) is given. Without it, the gains could come from general improvements in generation rather than prompt following.

Training details, data splits, and checkpoint selection criteria are also thin in the supplied text, which makes it hard to judge robustness. The paper is honest about these gaps.

This is for groups already running sequence models on synthesis planning who want a single architecture instead of one model per constraint type. A reader who needs a working prompt-conditioned generator with code will get something concrete to try. It is worth sending to peer review because the core idea is simple, the benchmarks are named, and the code is public; referees can ask for the compliance numbers directly.

Referee Report

2 major / 1 minor

Summary. The paper introduces Ariadne, a decoder-only transformer for direct retrosynthetic route generation that encodes the target molecule plus optional constraints (route depth, required starting materials) into a single prompt-completion sequence. On the RetroCast/PaRoutes mkt-cnv-160 family, a single 24-layer checkpoint shows Solv-0 gains of +13.7 (depth) and +31.2 (required-leaf) when the corresponding prompt fields are added; it also reports faster inference than DESP on required-leaf tasks and parity with DMS Explorer XL on standard reconstruction, while releasing training code but noting the absence of Tier-1–3 route checkers.

Significance. If the central claim holds, the work demonstrates that prompt conditioning can extend sequence-generation models to constrained retrosynthesis without bespoke search algorithms or multiple specialist models, offering a simpler interface for route-depth and starting-material constraints. The public release of code and scripts is a concrete strength that enables follow-up verification.

major comments (2)

[Abstract / Results] Abstract and results (headline claims on RetroCast/PaRoutes): the statement that the model 'follows' depth and required-starting-material prompts rests on downstream Solv-0/Top-10 improvements alone. No separate metric (e.g., fraction of routes whose depth ≤ prompted limit or that contain the required leaf, measured by an external checker) is reported, so it remains possible that gains arise from routes that violate the constraints.
[Experiments] Experiments section: training details (data splits, hyper-parameters, checkpoint selection procedure for the 24-layer model, and any statistical significance tests on the +13.7 / +31.2 point deltas) are not supplied, preventing assessment of whether the reported gains are robust or could be artifacts of post-hoc selection.

minor comments (1)

[Methods] Notation for prompt fields and route representation could be clarified with an explicit example sequence in the methods.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of evaluation and reproducibility. We address each major comment below and will revise the manuscript to incorporate additional details and analyses.

read point-by-point responses

Referee: [Abstract / Results] Abstract and results (headline claims on RetroCast/PaRoutes): the statement that the model 'follows' depth and required-starting-material prompts rests on downstream Solv-0/Top-10 improvements alone. No separate metric (e.g., fraction of routes whose depth ≤ prompted limit or that contain the required leaf, measured by an external checker) is reported, so it remains possible that gains arise from routes that violate the constraints.

Authors: We acknowledge that the evaluation relies on Solv-0 and Top-10 as aggregate success metrics under prompt conditioning, without an explicit post-hoc verification of constraint satisfaction on the generated routes. The model is trained on prompt-completion pairs where completions are consistent with the constraints present in the training data, but we agree that reporting the fraction of routes meeting the depth limit or containing the required leaf would strengthen the claim. In the revision we will add this analysis by parsing the generated route sequences (a lightweight check that does not require Tier-1–3 validation) and reporting the adherence rates for the prompted versus unprompted settings. revision: yes
Referee: [Experiments] Experiments section: training details (data splits, hyper-parameters, checkpoint selection procedure for the 24-layer model, and any statistical significance tests on the +13.7 / +31.2 point deltas) are not supplied, preventing assessment of whether the reported gains are robust or could be artifacts of post-hoc selection.

Authors: We agree these details should have been included. The revised Experiments section will specify the exact data splits from the RetroCast/PaRoutes mkt-cnv-160 family, the full hyper-parameter configuration for the 24-layer decoder-only model, and the checkpoint selection criterion (validation loss on a held-out set). The reported deltas come from the single selected checkpoint; we did not run multiple random seeds for statistical testing owing to training cost, and we will state this limitation explicitly while noting that the gains are reproducible from the released training scripts. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical gains measured on external benchmarks

full rationale

The paper's central claims consist of empirical performance deltas (Solv-0 gains of +13.7 and +31.2) obtained by adding prompt fields to an external benchmark family (RetroCast/PaRoutes mkt-cnv-160) and comparing against independent baselines (DESP, DMS Explorer XL, AiZynthFinder). No equations, fitted parameters, or self-citations are shown that reduce these deltas to quantities defined by the authors' own inputs or prior work. The derivation chain is therefore self-contained against external data and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

2 free parameters · 0 axioms · 0 invented entities

The approach rests on the standard transformer decoder architecture and training procedures from prior language-model literature; no new physical or chemical axioms are introduced.

free parameters (2)

24-layer model size
Chosen checkpoint size; value not derived from first principles.
training hyperparameters
Standard ML training choices whose values affect reported performance.

pith-pipeline@v0.9.1-grok · 5810 in / 1133 out tokens · 42692 ms · 2026-06-26T00:34:04.936530+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

71 extracted references · 28 canonical work pages · 1 internal anchor

[1]

Grand challenges for predictive modeling in small molecule drug discovery.ChemRxiv, 2026(0304), 2026

Connor W Coley, Pankaj Daga, Marco De Vivo, Willem Jespers, Ashutosh S Jogalekar, S Roy Kimura, Lucien Koenekoop, Anne-Grete Märtson, Timothy R Newhouse, Soumya Ray, Riccardo Saba- tini, David C Thompson, and Woody Sherman. Grand challenges for predictive modeling in small molecule drug discovery.ChemRxiv, 2026(0304), 2026. doi: 10.26434/chemrxiv.15000615...

work page doi:10.26434/chemrxiv.15000615/v1 2026
[2]

ERTL AND A

Peter Ertl and Ansgar Schuffenhauer. Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions.Journal of Cheminformatics, 1(1):8, 2009. ISSN 1758-2946. doi: 10.1186/1758-2946-1-8. URLhttps://doi.org/10.1186/1758-2946-1-8

work page doi:10.1186/1758-2946-1-8 2009
[3]

Coley, Luke Rogers, William H

Connor W. Coley, Luke Rogers, William H. Green, and Klavs F. Jensen. Scscore: Synthetic complexity learned from a reaction corpus.Journal of Chemical Information and Modeling, 58(2):252–261, 2018. doi: 10.1021/acs.jcim.7b00622. URL https://doi.org/10.1021/acs.jcim.7b00622. PMID: 29309147

work page doi:10.1021/acs.jcim.7b00622 2018
[4]

Retrosynthetic accessibility score (RAscore) – rapid machine learned synthesizability classification from AI driven retrosynthetic planning.Chemical Science, 12:3339 – 3349, 2020

Amol Thakkar, Veronika Chadimová, Esben Jannik Bjerrum, Ola Engkvist, and Jean-Louis Reymond. Retrosynthetic accessibility score (RAscore) – rapid machine learned synthesizability classification from AI driven retrosynthetic planning.Chemical Science, 12:3339 – 3349, 2020. URL https://api. semanticscholar.org/CorpusID:233621461

2020
[5]

Syba: Bayesian estimation of synthetic accessibility of organic compounds.Journal of Cheminformatics, 12(1):35, 2020

Milan V oršilák, Michał Koláˇr, Ivan ˇCmelo, and Daniel Svozil. Syba: Bayesian estimation of synthetic accessibility of organic compounds.Journal of Cheminformatics, 12(1):35, 2020. ISSN 1758-2946. doi: 10.1186/s13321-020-00439-2. URLhttps://doi.org/10.1186/s13321-020-00439-2

work page doi:10.1186/s13321-020-00439-2 2020
[6]

Integrating synthetic accessibility with AI-based generative drug design.Journal of Cheminformatics, 15, 2021

Maud Parrot, Hamza Tajmouati, Vinicius Barros Ribeiro da Silva, Brian Ross Atwood, Robin Fourcade, Yann Gaston-Mathé, Nicolas Do Huu, and Quentin Perron. Integrating synthetic accessibility with AI-based generative drug design.Journal of Cheminformatics, 15, 2021. URL https://api.semanticscholar. org/CorpusID:245417923

2021
[7]

The syntax of matter: Synthesis planning as the foundation of generative chemistry.ChemRxiv, 2026(0421), 2026

Anton Morgunov, Yu Shee, Alexander V Soudackov, and Victor S Batista. The syntax of matter: Synthesis planning as the foundation of generative chemistry.ChemRxiv, 2026(0421), 2026. doi: 10.26434/chemrxiv. 15001278/v3. URLhttps://chemrxiv.org/doi/abs/10.26434/chemrxiv.15001278/v3. 9

work page doi:10.26434/chemrxiv 2026
[8]

Michał Koziarski, Andrei Rekesh, Dmytro Shevchuk, Almer van der Sloot, Piotr Gai ´nski, Yoshua Bengio, Cheng-Hao Liu, Mike Tyers, and Robert A. Batey. RGFN: Synthesizable molecular gen- eration using GFlowNets. InAdvances in Neural Information Processing Systems, volume 37, pages 46908–46955. Neural Information Processing Systems Foundation, Inc., 2024. d...

2024
[9]

Zygimantas Jocys, Zhanxing Zhu, Henriette M. G. Willems, and Katayoun Farrahi. Synthformer: Equiv- ariant pharmacophore-based generation of synthesizable molecules for ligand-based drug design.Ar- tificial Intelligence in the Life Sciences, 9:100148, 2026. doi: 10.1016/j.ailsci.2025.100148. URL https://doi.org/10.1016/j.ailsci.2025.100148

work page doi:10.1016/j.ailsci.2025.100148 2026
[10]

Generative flows on synthetic pathway for drug design, 2025

Seonghwan Seo, Minsu Kim, Tony Shen, Martin Ester, Jinkyoo Park, Sungsoo Ahn, and Woo Youn Kim. Generative flows on synthetic pathway for drug design, 2025. URL https://arxiv.org/abs/2410. 04542

2025
[11]

Shitong Luo and Connor W. Coley. Efficient and programmable exploration of synthesizable chemical space, 2025. URLhttps://arxiv.org/abs/2512.00384

arXiv 2025
[12]

E. J. Corey and W. Todd Wipke. Computer-assisted design of complex organic syntheses: Pathways for molecular synthesis can be devised with a computer and equipment for graphical communication. Science, 166(3902):178–192, 1969. ISSN 0036-8075, 1095-9203. doi: 10.1126/science.166.3902.178. URLhttps://www.science.org/doi/10.1126/science.166.3902.178

work page doi:10.1126/science.166.3902.178 1969
[13]

Marwin H. S. Segler, Mike Preuss, and Mark P. Waller. Planning chemical syntheses with deep neural networks and symbolic AI.Nature, 555(7698):604–610, 2018. ISSN 1476-4687. doi: 10.1038/nature25978. URLhttps://doi.org/10.1038/nature25978

work page doi:10.1038/nature25978 2018
[14]

Depth-first proof-number search with heuristic edge cost and application to chemical synthesis planning

Akihiro Kishimoto, Beat Buesser, Bei Chen, and Adi Botea. Depth-first proof-number search with heuristic edge cost and application to chemical synthesis planning. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. URL htt...

2019
[15]

Schreck, Connor W

John S. Schreck, Connor W. Coley, and Kyle J. M. Bishop. Learning retrosynthetic planning through simulated experience.ACS Central Science, 5(6):970–981, 2019. doi: 10.1021/acscentsci.9b00055

work page doi:10.1021/acscentsci.9b00055 2019
[16]

Coley, Yiming Mo, Regina Barzilay, and Klavs F

Xiaoxue Wang, Yujie Qian, Hanyu Gao, Connor W. Coley, Yiming Mo, Regina Barzilay, and Klavs F. Jensen. Towards efficient discovery of green synthetic pathways with monte carlo tree search and reinforcement learning.Chem. Sci., 11:10959–10972, 2020. doi: 10.1039/D0SC04184J. URL http: //dx.doi.org/10.1039/D0SC04184J

work page doi:10.1039/d0sc04184j 2020
[17]

Retro*: Learning retrosynthetic planning with neural guided A* search

Binghong Chen, Chengtao Li, Hanjun Dai, and Le Song. Retro*: Learning retrosynthetic planning with neural guided A* search. InProceedings of the 37th International Conference on Machine Learning, volume 119 ofProceedings of Machine Learning Research, pages 1608–1616. PMLR, 2020. URL https://proceedings.mlr.press/v119/chen20k.html

2020
[18]

AiZynthFinder: a fast, robust and flexible open-source software for retrosynthetic planning

Samuel Genheden, Amol Thakkar, Veronika Chadimová, Jean-Louis Reymond, Ola Engkvist, and Esben Bjerrum. AiZynthFinder: a fast, robust and flexible open-source software for retrosynthetic planning. Journal of Cheminformatics, 12(1):70, 2020. ISSN 1758-2946. doi: 10.1186/s13321-020-00472-1. URL https://doi.org/10.1186/s13321-020-00472-1

work page doi:10.1186/s13321-020-00472-1 2020
[19]

Grasp: Navigating retrosyn- thetic planning with goal-driven policy

Yemin Yu, Ying Wei, Kun Kuang, Zhengxing Huang, Huaxiu Yao, and Fei Wu. Grasp: Navigating retrosyn- thetic planning with goal-driven policy. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors,Advances in Neural Information Processing Systems, volume 35, pages 10257–10268. Cur- ran Associates, Inc., 2022. URL https://proceedings....

2022
[20]

Retrograph: Retrosynthetic planning with graph search

Shufang Xie, Rui Yan, Peng Han, Yingce Xia, Lijun Wu, Chenjuan Guo, Bin Yang, and Tao Qin. Retrograph: Retrosynthetic planning with graph search. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD ’22, pages 2120–2129, New York, NY , USA,
[21]

ISBN 9781450393850

Association for Computing Machinery. ISBN 9781450393850. doi: 10.1145/3534678.3539446. URLhttps://doi.org/10.1145/3534678.3539446

work page doi:10.1145/3534678.3539446
[22]

Retrosynthetic planning with experience-guided monte carlo tree search.Communications Chemistry, 6(1):120, 2023

Siqi Hong, Hankz Hankui Zhuo, Kebing Jin, Guang Shao, and Zhanwen Zhou. Retrosynthetic planning with experience-guided monte carlo tree search.Communications Chemistry, 6(1):120, 2023. doi: 10.1038/s42004-023-00911-8. URLhttps://doi.org/10.1038/s42004-023-00911-8. 10

work page doi:10.1038/s42004-023-00911-8 2023
[23]

Guoqing Liu, Di Xue, Shufang Xie, Yingce Xia, Austin Tripp, Krzysztof Maziarz, Marwin H. S. Segler, Tao Qin, Zongzhang Zhang, and Tie-Yan Liu. Retrosynthetic planning with dual value networks. In International Conference on Machine Learning, 2023. URL https://api.semanticscholar.org/ CorpusID:256416110

2023
[24]

Evolutionary retrosynthetic route planning [research frontier].IEEE Computational Intelligence Magazine, 19:58–72, 2023

Yan Zhang, Xiao He, Shuanhu Gao, Aimin Zhou, and Hao Hao. Evolutionary retrosynthetic route planning [research frontier].IEEE Computational Intelligence Magazine, 19:58–72, 2023. URL https: //api.semanticscholar.org/CorpusID:271115363

2023
[25]

Efficient retrosynthetic planning with MCTS exploration en- hanced A* search.Communications Chemistry, 7, 2024

Dengwei Zhao, Shikui Tu, and Lei Xu. Efficient retrosynthetic planning with MCTS exploration en- hanced A* search.Communications Chemistry, 7, 2024. URL https://api.semanticscholar.org/ CorpusID:268252759

2024
[26]

Retro- fallback: retrosynthetic planning in an uncertain world, 2024

Austin Tripp, Krzysztof Maziarz, Sarah Lewis, Marwin Segler, and José Miguel Hernández-Lobato. Retro- fallback: retrosynthetic planning in an uncertain world, 2024. URL https://arxiv.org/abs/2310. 09270

2024
[27]

Retrosynthesis zero: Self-improving global synthesis planning using reinforcement learning.Journal of chemical theory and computation, 2024

Jiasheng Guo, Chenning Yu, Kenan Li, Yijian Zhang, Guoqiang Wang, Shuhua Li, and Hao Dong. Retrosynthesis zero: Self-improving global synthesis planning using reinforcement learning.Journal of chemical theory and computation, 2024. URL https://api.semanticscholar.org/CorpusID: 269771006

2024
[28]

Davies, Kristian T Spoerer, and Jonathan D

Ton M Blackshaw, Joseph C. Davies, Kristian T Spoerer, and Jonathan D. Hirst. Enhancing Monte Carlo Tree Search for retrosynthesis.Journal of Chemical Information and Modeling, 65:6537 – 6546, 2025. URLhttps://api.semanticscholar.org/CorpusID:279328860

2025
[29]

A data-driven group retrosyn- thesis planning model inspired by neurosymbolic programming.Nature Communications, 16(1):192, 2025

Xuefeng Zhang, Haowei Lin, Muhan Zhang, Yuan Zhou, and Jianzhu Ma. A data-driven group retrosyn- thesis planning model inspired by neurosymbolic programming.Nature Communications, 16(1):192, 2025. doi: 10.1038/s41467-024-55374-9. URLhttps://doi.org/10.1038/s41467-024-55374-9

work page doi:10.1038/s41467-024-55374-9 2025
[30]

Joung, Kevin Yu, Zhengkai Tu, G

Jihye Roh, Joonyoung F. Joung, Kevin Yu, Zhengkai Tu, G. Logan Bartholomew, Omar A. Santiago-Reyes, Mun Hong Fong, Richmond Sarpong, Sarah E. Reisman, and Connor W. Coley. Higher-level strategies for computer-aided retrosynthesis.ACS Central Science, 12(3):345–357, 2026. doi: 10.1021/acscentsci. 5c02014. URLhttps://doi.org/10.1021/acscentsci.5c02014

work page doi:10.1021/acscentsci 2026
[31]

Synplanner: An end-to-end tool for synthesis planning.Journal of Chemical Information and Modeling, 65(1):15–21, 2025

Tagir Akhmetshin, Dmitry Zankov, Philippe Gantzer, Dmitry Babadeev, Anna Pinigina, Timur Madzhidov, and Alexandre Varnek. Synplanner: An end-to-end tool for synthesis planning.Journal of Chemical Information and Modeling, 65(1):15–21, 2025. doi: 10.1021/acs.jcim.4c02004

work page doi:10.1021/acs.jcim.4c02004 2025
[32]

Retrosynthesis planning via worst-path policy optimisation in tree-structured mdps, 2025

Mianchu Wang and Giovanni Montana. Retrosynthesis planning via worst-path policy optimisation in tree-structured mdps, 2025. URLhttps://arxiv.org/abs/2509.10504

arXiv 2025
[33]

AI-driven synthetic route design incorporated with retrosynthesis knowledge.Journal of Chemical Information and Modeling, 62(6):1357–1367, 2022

Shoichi Ishida, Kei Terayama, Ryosuke Kojima, Kiyosei Takasu, and Yasushi Okuno. AI-driven synthetic route design incorporated with retrosynthesis knowledge.Journal of Chemical Information and Modeling, 62(6):1357–1367, 2022. doi: 10.1021/acs.jcim.1c01074

work page doi:10.1021/acs.jcim.1c01074 2022
[34]

Comparing search algorithms on the retrosynthesis problem

Milo Roucairol and Tristan Cazenave. Comparing search algorithms on the retrosynthesis problem. Molecular Informatics, 43, 2024. URL https://api.semanticscholar.org/CorpusID:253882602

2024
[35]

Multimodal large language models for inverse molecular design with retrosynthetic planning, 2024

Gang Liu, Michael Sun, Wojciech Matusik, Meng Jiang, and Jie Chen. Multimodal large language models for inverse molecular design with retrosynthetic planning, 2024. URL https://arxiv.org/abs/2410. 04223

2024
[36]

Kevin Yu, Jihye Roh, Ziang Li, Wenhao Gao, Runzhong Wang, and Connor W. Coley. Double-ended synthesis planning with goal-constrained bidirectional search. InAdvances in Neural Information Pro- cessing Systems, volume 37, pages 112919–112949. Neural Information Processing Systems Foundation, Inc., 2024. doi: 10.52202/079017-3588. URL https://proceedings.ne...

work page doi:10.52202/079017-3588 2024
[37]

Chemist-aligned retrosynthesis by ensembling diverse inductive bias models, 2025

Krzysztof Maziarz, Guoqing Liu, Hubert Misztela, Austin Tripp, Junren Li, Aleksei Kornev, Piotr Gai´nski, Holger Hoefling, Mike Fortunato, Rishi Gupta, and Marwin Segler. Chemist-aligned retrosynthesis by ensembling diverse inductive bias models, 2025. URLhttps://arxiv.org/abs/2412.05269

arXiv 2025
[38]

Baker, Daniel Adu-Ampratwum, Reza Averly, Botao Yu, Huan Sun, and Xia Ning

Frazier N. Baker, Daniel Adu-Ampratwum, Reza Averly, Botao Yu, Huan Sun, and Xia Ning. LARC: Towards human-level constrained retrosynthesis planning through an agentic framework. InProceedings of AI for Accelerated Research Symposium, volume 3 ofEPiC Series in Technology, pages 153–176. EasyChair, 2026. doi: 10.29007/z3hb. URLhttps://easychair.org/publica...

work page doi:10.29007/z3hb 2026
[39]

Bran, Zlatko Jonˇcev, and Philippe Schwaller

Nguyen Xuan-Vu, Daniel Armstrong, Milena Wehrbach, Andres M. Bran, Zlatko Jonˇcev, and Philippe Schwaller. Synthelite: Chemist-aligned and feasibility-aware synthesis planning with LLMs, 2025. URL https://arxiv.org/abs/2512.16424

arXiv 2025
[40]

AOT*: Efficient synthesis planning via LLM-empowered AND-OR tree search, 2025

Xiaozhuang Song, Xuanhao Pan, Xinjian Zhao, Hangting Ye, Shufei Zhang, Jian Tang, and Tianshu Yu. AOT*: Efficient synthesis planning via LLM-empowered AND-OR tree search, 2025. URL https: //arxiv.org/abs/2509.20988

arXiv 2025
[41]

Diverse and feasible retrosynthesis using GFlowNets.Information Sciences, 714:122194, 2025

Piotr Gai ´nski, Michał Koziarski, Krzysztof Maziarz, Marwin Segler, Jacek Tabor, and Marek ´Smieja. Diverse and feasible retrosynthesis using GFlowNets.Information Sciences, 714:122194, 2025. doi: 10.1016/j.ins.2025.122194. URLhttps://doi.org/10.1016/j.ins.2025.122194

work page doi:10.1016/j.ins.2025.122194 2025
[42]

Nair, Rico Häuselmann, Riccardo Pisoni, Costas Bekas, Anna Iuliano, and Teodoro Laino

Philippe Schwaller, Riccardo Petraglia, Valerio Zullo, Vishnu H. Nair, Rico Häuselmann, Riccardo Pisoni, Costas Bekas, Anna Iuliano, and Teodoro Laino. Predicting retrosynthetic pathways using transformer- based models and a hyper-graph exploration strategy.Chemical Science, 11:3316 – 3325, 2020. URL https://api.semanticscholar.org/CorpusID:216332642

2020
[43]

Automatic retrosynthetic route planning using template-free models.Chemical Science, 11:3355 – 3364, 2020

Kangjie Lin, Youjun Xu, Jianfeng Pei, and Luhua Lai. Automatic retrosynthetic route planning using template-free models.Chemical Science, 11:3355 – 3364, 2020. URL https://api.semanticscholar. org/CorpusID:268816571

2020
[44]

Multistep retrosynthesis combining a disconnection aware triple transformer loop with a route penalty score guided tree search.Chemical Science, 14:9959 – 9969, 2023

David Kreutter and Jean-Louis Reymond. Multistep retrosynthesis combining a disconnection aware triple transformer loop with a route penalty score guided tree search.Chemical Science, 14:9959 – 9969, 2023. URLhttps://api.semanticscholar.org/CorpusID:261488982

2023
[45]

Yu Shee, Anton Morgunov, Haote Li, and Victor S. Batista. DirectMultiStep: Direct route generation for multistep retrosynthesis.Journal of Chemical Information and Modeling, 65(8):3903–3914, 2025. doi: 10.1021/acs.jcim.4c01982

work page doi:10.1021/acs.jcim.4c01982 2025
[46]

Fast and scalable retrosynthetic planning with a transformer neural network and speculative beam search

Natalia Andronova, Mikhail Andronov, Jürgen Schmidhuber, Michael Wand, and Djork-Arné Clevert. Fast and scalable retrosynthetic planning with a transformer neural network and speculative beam search. Digital Discovery, 5:1783–1793, 2026. doi: 10.1039/D5DD00573F. URL http://dx.doi.org/10. 1039/D5DD00573F

work page doi:10.1039/d5dd00573f 2026
[47]

Cavanagh, Yingze Wang, Jacob M

Kunyang Sun, Dorian Bagni, Joseph M. Cavanagh, Yingze Wang, Jacob M. Sawyer, Bo Zhou, Andrew Gritsevskiy, Oufan Zhang, and Teresa Head-Gordon. Synllama: Generating synthesizable molecules and their analogs with large language models.ACS Central Science, 11(11):2108–2120, 2025. doi: 10.1021/acscentsci.5c01285

work page doi:10.1021/acscentsci.5c01285 2025
[48]

TempRe: Template generation for single and direct multi-step retrosynthesis, 2025

Nguyen Xuan-Vu, Daniel P Armstrong, Zlatko Jon ˇcev, and Philippe Schwaller. TempRe: Template generation for single and direct multi-step retrosynthesis, 2025. URL https://arxiv.org/abs/2507. 21762

2025
[49]

LLM-augmented chemical synthesis and design decision programs, 2025

Haorui Wang, Jeff Guo, Lingkai Kong, Rampi Ramprasad, Philippe Schwaller, Yuanqi Du, and Chao Zhang. LLM-augmented chemical synthesis and design decision programs, 2025. URL https://arxiv. org/abs/2505.07027

Pith/arXiv arXiv 2025
[50]

Retrosynformer: planning multi-step chemical synthesis routes via a decision transformer.Digital Discovery, 5:348–362, 2026

Emma Granqvist, Rocío Mercado, and Samuel Genheden. Retrosynformer: planning multi-step chemical synthesis routes via a decision transformer.Digital Discovery, 5:348–362, 2026. doi: 10.1039/D5DD00153F. URLhttp://dx.doi.org/10.1039/D5DD00153F

work page doi:10.1039/d5dd00153f 2026
[51]

PaRoutes: towards a framework for benchmarking retrosynthesis route predictions.Digital Discovery, 1:527–539, 2022

Samuel Genheden and Esben Bjerrum. PaRoutes: towards a framework for benchmarking retrosynthesis route predictions.Digital Discovery, 1:527–539, 2022. doi: 10.1039/D2DD00015F. URL http: //dx.doi.org/10.1039/D2DD00015F

work page doi:10.1039/d2dd00015f 2022
[52]

Anton Morgunov and Victor S. Batista. Procrustean bed for AI-driven retrosynthesis: A unified framework for reproducible evaluation, 2025. URLhttps://arxiv.org/abs/2512.07079

arXiv 2025
[53]

Choure, Mun Hong Fong, Jihye Roh, Itai Levin, Kevin Yu, Joonyoung F

Zhengkai Tu, Sourabh J. Choure, Mun Hong Fong, Jihye Roh, Itai Levin, Kevin Yu, Joonyoung F. Joung, Nathan Morgan, Shih-Cheng Li, Xiaoqi Sun, Huiqian Lin, Mark Murnin, Jordan P. Liles, Thomas J. Struble, Michael E. Fortunato, Mengjie Liu, William H. Green, Klavs F. Jensen, and Connor W. Coley. ASKCOS: Open-source, data-driven synthesis planning.Accounts o...

work page doi:10.1021/acs.accounts.5c00155 2025
[54]

Unified language model pre-training for natural language understanding and generation

Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xiaodong Liu, Yu Wang, Jianfeng Gao, Ming Zhou, and Hsiao-Wuen Hon. Unified language model pre-training for natural language understanding and generation. InAdvances in Neural Information Processing Systems, volume 32, 2019. URL https://proceedings. neurips.cc/paper/2019/hash/c20bb2d9a50d5ac1f713f8b34d9aac5a-Abstr...

2019
[55]

Richard S. Sutton. The bitter lesson, 2019. URL http://www.incompleteideas.net/IncIdeas/ BitterLesson.html

2019
[56]

Attention is all you need

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In I. Guyon, U. V on Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc., 2017. URL htt...

2017
[57]

doi: https://doi.org/10.1016/j.neucom

Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024. doi: 10.1016/j.neucom. 2023.127063. URLhttps://doi.org/10.1016/j.neucom.2023.127063

work page doi:10.1016/j.neucom 2024
[58]

Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization, 2016. URL https: //arxiv.org/abs/1607.06450

Pith/arXiv arXiv 2016
[59]

Root mean square layer normalization

Biao Zhang and Rico Sennrich. Root mean square layer normalization. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. URL https://proceedings.neurips.cc/paper_ files/paper/2019/file/1e8a19426224ca89e83cef47f1e7f53b-Paper.pdf

2019
[60]

On layer normalization in the transformer architecture

Ruibin Xiong, Yunchang Yang, Di He, Kai Zheng, Shuxin Zheng, Chen Xing, Huishuai Zhang, Yanyan Lan, Liwei Wang, and Tie-Yan Liu. On layer normalization in the transformer architecture. InPro- ceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event, Proceedings of Machine Learning Research, pages 10524–1...

2020
[61]

Gaussian error linear units (gelus), 2016

Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus), 2016. URL https://arxiv.org/ abs/1606.08415

Pith/arXiv arXiv 2016
[62]

Glu variants improve transformer, 2020

Noam Shazeer. Glu variants improve transformer, 2020. URL https://arxiv.org/abs/2002.05202

Pith/arXiv arXiv 2020
[63]

Llama: Open and efficient foundation language models, 2023

Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. Llama: Open and efficient foundation language models, 2023. URL https://arxiv.org/abs/2302.13971

Pith/arXiv arXiv 2023
[64]

Decoupled weight decay regularization

Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net, 2019. URLhttps://openreview.net/forum?id=Bkg6RiCqY7

2019
[65]

Muon: An optimizer for hidden layers in neural networks, 2024

Keller Jordan, Yuchen Jin, Vlado Boza, Jiacheng You, Franz Cesista, Laker Newhouse, and Jeremy Bern- stein. Muon: An optimizer for hidden layers in neural networks, 2024. URL https://kellerjordan. github.io/posts/muon/

2024
[66]

Muon is scalable for LLM training, 2025

Jingyuan Liu, Jianlin Su, Xingcheng Yao, Zhejun Jiang, Guokun Lai, Yulun Du, Yidao Qin, Weixin Xu, Enzhe Lu, Junjie Yan, Yanru Chen, Huabin Zheng, Yibo Liu, Shaowei Liu, Bohong Yin, Weiran He, Han Zhu, Yuzhi Wang, Jianzhou Wang, Mengnan Dong, Zheng Zhang, Yongsheng Kang, Hao Zhang, Xinran Xu, Yutao Zhang, Yuxin Wu, Xinyu Zhou, and Zhilin Yang. Muon is sca...

Pith/arXiv arXiv 2025
[67]

e3nn: Euclidean neural networks,

Reiner Pope, Sholto Douglas, Aakanksha Chowdhery, Jacob Devlin, James Bradbury, Anselm Levskaya, Jonathan Heek, Kefan Xiao, Shivani Agrawal, and Jeff Dean. Efficiently scaling transformer inference. CoRR, abs/2211.05102, 2022. doi: 10.48550/ARXIV .2211.05102. URL https://doi.org/10.48550/ arXiv.2211.05102. 13 Supporting Information The Supporting Informat...

work page internal anchor Pith review doi:10.48550/arxiv 2022
[68]

Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered

Top-10 CI gives the 95% bootstrap interval. Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered. Paired CI rows compare Ariadne with AiZynthFinder MCTS: intervals containing zero indicate no significant difference, positive intervals favor Ariadne, and n...
[69]

Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered

Top-10 CI gives the 95% bootstrap interval. Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered. Paired CI rows compare Ariadne with AiZynthFinder MCTS: intervals containing zero indicate no significant difference, positive intervals favor Ariadne, and n...
[70]

Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered

Top-10 CI gives the 95% bootstrap interval. Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered. Paired CI rows compare Ariadne with AiZynthFinder MCTS: intervals containing zero indicate no significant difference, positive intervals favor Ariadne, and n...
[71]

Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered

Top-10 CI gives the 95% bootstrap interval. Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered. Paired CI rows compare Ariadne with AiZynthFinder MCTS: intervals containing zero indicate no significant difference, positive intervals favor Ariadne, and n...

[1] [1]

Grand challenges for predictive modeling in small molecule drug discovery.ChemRxiv, 2026(0304), 2026

Connor W Coley, Pankaj Daga, Marco De Vivo, Willem Jespers, Ashutosh S Jogalekar, S Roy Kimura, Lucien Koenekoop, Anne-Grete Märtson, Timothy R Newhouse, Soumya Ray, Riccardo Saba- tini, David C Thompson, and Woody Sherman. Grand challenges for predictive modeling in small molecule drug discovery.ChemRxiv, 2026(0304), 2026. doi: 10.26434/chemrxiv.15000615...

work page doi:10.26434/chemrxiv.15000615/v1 2026

[2] [2]

ERTL AND A

Peter Ertl and Ansgar Schuffenhauer. Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions.Journal of Cheminformatics, 1(1):8, 2009. ISSN 1758-2946. doi: 10.1186/1758-2946-1-8. URLhttps://doi.org/10.1186/1758-2946-1-8

work page doi:10.1186/1758-2946-1-8 2009

[3] [3]

Coley, Luke Rogers, William H

Connor W. Coley, Luke Rogers, William H. Green, and Klavs F. Jensen. Scscore: Synthetic complexity learned from a reaction corpus.Journal of Chemical Information and Modeling, 58(2):252–261, 2018. doi: 10.1021/acs.jcim.7b00622. URL https://doi.org/10.1021/acs.jcim.7b00622. PMID: 29309147

work page doi:10.1021/acs.jcim.7b00622 2018

[4] [4]

Retrosynthetic accessibility score (RAscore) – rapid machine learned synthesizability classification from AI driven retrosynthetic planning.Chemical Science, 12:3339 – 3349, 2020

Amol Thakkar, Veronika Chadimová, Esben Jannik Bjerrum, Ola Engkvist, and Jean-Louis Reymond. Retrosynthetic accessibility score (RAscore) – rapid machine learned synthesizability classification from AI driven retrosynthetic planning.Chemical Science, 12:3339 – 3349, 2020. URL https://api. semanticscholar.org/CorpusID:233621461

2020

[5] [5]

Syba: Bayesian estimation of synthetic accessibility of organic compounds.Journal of Cheminformatics, 12(1):35, 2020

Milan V oršilák, Michał Koláˇr, Ivan ˇCmelo, and Daniel Svozil. Syba: Bayesian estimation of synthetic accessibility of organic compounds.Journal of Cheminformatics, 12(1):35, 2020. ISSN 1758-2946. doi: 10.1186/s13321-020-00439-2. URLhttps://doi.org/10.1186/s13321-020-00439-2

work page doi:10.1186/s13321-020-00439-2 2020

[6] [6]

Integrating synthetic accessibility with AI-based generative drug design.Journal of Cheminformatics, 15, 2021

Maud Parrot, Hamza Tajmouati, Vinicius Barros Ribeiro da Silva, Brian Ross Atwood, Robin Fourcade, Yann Gaston-Mathé, Nicolas Do Huu, and Quentin Perron. Integrating synthetic accessibility with AI-based generative drug design.Journal of Cheminformatics, 15, 2021. URL https://api.semanticscholar. org/CorpusID:245417923

2021

[7] [7]

The syntax of matter: Synthesis planning as the foundation of generative chemistry.ChemRxiv, 2026(0421), 2026

Anton Morgunov, Yu Shee, Alexander V Soudackov, and Victor S Batista. The syntax of matter: Synthesis planning as the foundation of generative chemistry.ChemRxiv, 2026(0421), 2026. doi: 10.26434/chemrxiv. 15001278/v3. URLhttps://chemrxiv.org/doi/abs/10.26434/chemrxiv.15001278/v3. 9

work page doi:10.26434/chemrxiv 2026

[8] [8]

Michał Koziarski, Andrei Rekesh, Dmytro Shevchuk, Almer van der Sloot, Piotr Gai ´nski, Yoshua Bengio, Cheng-Hao Liu, Mike Tyers, and Robert A. Batey. RGFN: Synthesizable molecular gen- eration using GFlowNets. InAdvances in Neural Information Processing Systems, volume 37, pages 46908–46955. Neural Information Processing Systems Foundation, Inc., 2024. d...

2024

[9] [9]

Zygimantas Jocys, Zhanxing Zhu, Henriette M. G. Willems, and Katayoun Farrahi. Synthformer: Equiv- ariant pharmacophore-based generation of synthesizable molecules for ligand-based drug design.Ar- tificial Intelligence in the Life Sciences, 9:100148, 2026. doi: 10.1016/j.ailsci.2025.100148. URL https://doi.org/10.1016/j.ailsci.2025.100148

work page doi:10.1016/j.ailsci.2025.100148 2026

[10] [10]

Generative flows on synthetic pathway for drug design, 2025

Seonghwan Seo, Minsu Kim, Tony Shen, Martin Ester, Jinkyoo Park, Sungsoo Ahn, and Woo Youn Kim. Generative flows on synthetic pathway for drug design, 2025. URL https://arxiv.org/abs/2410. 04542

2025

[11] [11]

Shitong Luo and Connor W. Coley. Efficient and programmable exploration of synthesizable chemical space, 2025. URLhttps://arxiv.org/abs/2512.00384

arXiv 2025

[12] [12]

E. J. Corey and W. Todd Wipke. Computer-assisted design of complex organic syntheses: Pathways for molecular synthesis can be devised with a computer and equipment for graphical communication. Science, 166(3902):178–192, 1969. ISSN 0036-8075, 1095-9203. doi: 10.1126/science.166.3902.178. URLhttps://www.science.org/doi/10.1126/science.166.3902.178

work page doi:10.1126/science.166.3902.178 1969

[13] [13]

Marwin H. S. Segler, Mike Preuss, and Mark P. Waller. Planning chemical syntheses with deep neural networks and symbolic AI.Nature, 555(7698):604–610, 2018. ISSN 1476-4687. doi: 10.1038/nature25978. URLhttps://doi.org/10.1038/nature25978

work page doi:10.1038/nature25978 2018

[14] [14]

Depth-first proof-number search with heuristic edge cost and application to chemical synthesis planning

Akihiro Kishimoto, Beat Buesser, Bei Chen, and Adi Botea. Depth-first proof-number search with heuristic edge cost and application to chemical synthesis planning. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. URL htt...

2019

[15] [15]

Schreck, Connor W

John S. Schreck, Connor W. Coley, and Kyle J. M. Bishop. Learning retrosynthetic planning through simulated experience.ACS Central Science, 5(6):970–981, 2019. doi: 10.1021/acscentsci.9b00055

work page doi:10.1021/acscentsci.9b00055 2019

[16] [16]

Coley, Yiming Mo, Regina Barzilay, and Klavs F

Xiaoxue Wang, Yujie Qian, Hanyu Gao, Connor W. Coley, Yiming Mo, Regina Barzilay, and Klavs F. Jensen. Towards efficient discovery of green synthetic pathways with monte carlo tree search and reinforcement learning.Chem. Sci., 11:10959–10972, 2020. doi: 10.1039/D0SC04184J. URL http: //dx.doi.org/10.1039/D0SC04184J

work page doi:10.1039/d0sc04184j 2020

[17] [17]

Retro*: Learning retrosynthetic planning with neural guided A* search

Binghong Chen, Chengtao Li, Hanjun Dai, and Le Song. Retro*: Learning retrosynthetic planning with neural guided A* search. InProceedings of the 37th International Conference on Machine Learning, volume 119 ofProceedings of Machine Learning Research, pages 1608–1616. PMLR, 2020. URL https://proceedings.mlr.press/v119/chen20k.html

2020

[18] [18]

AiZynthFinder: a fast, robust and flexible open-source software for retrosynthetic planning

Samuel Genheden, Amol Thakkar, Veronika Chadimová, Jean-Louis Reymond, Ola Engkvist, and Esben Bjerrum. AiZynthFinder: a fast, robust and flexible open-source software for retrosynthetic planning. Journal of Cheminformatics, 12(1):70, 2020. ISSN 1758-2946. doi: 10.1186/s13321-020-00472-1. URL https://doi.org/10.1186/s13321-020-00472-1

work page doi:10.1186/s13321-020-00472-1 2020

[19] [19]

Grasp: Navigating retrosyn- thetic planning with goal-driven policy

Yemin Yu, Ying Wei, Kun Kuang, Zhengxing Huang, Huaxiu Yao, and Fei Wu. Grasp: Navigating retrosyn- thetic planning with goal-driven policy. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors,Advances in Neural Information Processing Systems, volume 35, pages 10257–10268. Cur- ran Associates, Inc., 2022. URL https://proceedings....

2022

[20] [20]

Retrograph: Retrosynthetic planning with graph search

Shufang Xie, Rui Yan, Peng Han, Yingce Xia, Lijun Wu, Chenjuan Guo, Bin Yang, and Tao Qin. Retrograph: Retrosynthetic planning with graph search. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD ’22, pages 2120–2129, New York, NY , USA,

[21] [21]

ISBN 9781450393850

Association for Computing Machinery. ISBN 9781450393850. doi: 10.1145/3534678.3539446. URLhttps://doi.org/10.1145/3534678.3539446

work page doi:10.1145/3534678.3539446

[22] [22]

Retrosynthetic planning with experience-guided monte carlo tree search.Communications Chemistry, 6(1):120, 2023

Siqi Hong, Hankz Hankui Zhuo, Kebing Jin, Guang Shao, and Zhanwen Zhou. Retrosynthetic planning with experience-guided monte carlo tree search.Communications Chemistry, 6(1):120, 2023. doi: 10.1038/s42004-023-00911-8. URLhttps://doi.org/10.1038/s42004-023-00911-8. 10

work page doi:10.1038/s42004-023-00911-8 2023

[23] [23]

Guoqing Liu, Di Xue, Shufang Xie, Yingce Xia, Austin Tripp, Krzysztof Maziarz, Marwin H. S. Segler, Tao Qin, Zongzhang Zhang, and Tie-Yan Liu. Retrosynthetic planning with dual value networks. In International Conference on Machine Learning, 2023. URL https://api.semanticscholar.org/ CorpusID:256416110

2023

[24] [24]

Evolutionary retrosynthetic route planning [research frontier].IEEE Computational Intelligence Magazine, 19:58–72, 2023

Yan Zhang, Xiao He, Shuanhu Gao, Aimin Zhou, and Hao Hao. Evolutionary retrosynthetic route planning [research frontier].IEEE Computational Intelligence Magazine, 19:58–72, 2023. URL https: //api.semanticscholar.org/CorpusID:271115363

2023

[25] [25]

Efficient retrosynthetic planning with MCTS exploration en- hanced A* search.Communications Chemistry, 7, 2024

Dengwei Zhao, Shikui Tu, and Lei Xu. Efficient retrosynthetic planning with MCTS exploration en- hanced A* search.Communications Chemistry, 7, 2024. URL https://api.semanticscholar.org/ CorpusID:268252759

2024

[26] [26]

Retro- fallback: retrosynthetic planning in an uncertain world, 2024

Austin Tripp, Krzysztof Maziarz, Sarah Lewis, Marwin Segler, and José Miguel Hernández-Lobato. Retro- fallback: retrosynthetic planning in an uncertain world, 2024. URL https://arxiv.org/abs/2310. 09270

2024

[27] [27]

Retrosynthesis zero: Self-improving global synthesis planning using reinforcement learning.Journal of chemical theory and computation, 2024

Jiasheng Guo, Chenning Yu, Kenan Li, Yijian Zhang, Guoqiang Wang, Shuhua Li, and Hao Dong. Retrosynthesis zero: Self-improving global synthesis planning using reinforcement learning.Journal of chemical theory and computation, 2024. URL https://api.semanticscholar.org/CorpusID: 269771006

2024

[28] [28]

Davies, Kristian T Spoerer, and Jonathan D

Ton M Blackshaw, Joseph C. Davies, Kristian T Spoerer, and Jonathan D. Hirst. Enhancing Monte Carlo Tree Search for retrosynthesis.Journal of Chemical Information and Modeling, 65:6537 – 6546, 2025. URLhttps://api.semanticscholar.org/CorpusID:279328860

2025

[29] [29]

A data-driven group retrosyn- thesis planning model inspired by neurosymbolic programming.Nature Communications, 16(1):192, 2025

Xuefeng Zhang, Haowei Lin, Muhan Zhang, Yuan Zhou, and Jianzhu Ma. A data-driven group retrosyn- thesis planning model inspired by neurosymbolic programming.Nature Communications, 16(1):192, 2025. doi: 10.1038/s41467-024-55374-9. URLhttps://doi.org/10.1038/s41467-024-55374-9

work page doi:10.1038/s41467-024-55374-9 2025

[30] [30]

Joung, Kevin Yu, Zhengkai Tu, G

Jihye Roh, Joonyoung F. Joung, Kevin Yu, Zhengkai Tu, G. Logan Bartholomew, Omar A. Santiago-Reyes, Mun Hong Fong, Richmond Sarpong, Sarah E. Reisman, and Connor W. Coley. Higher-level strategies for computer-aided retrosynthesis.ACS Central Science, 12(3):345–357, 2026. doi: 10.1021/acscentsci. 5c02014. URLhttps://doi.org/10.1021/acscentsci.5c02014

work page doi:10.1021/acscentsci 2026

[31] [31]

Synplanner: An end-to-end tool for synthesis planning.Journal of Chemical Information and Modeling, 65(1):15–21, 2025

Tagir Akhmetshin, Dmitry Zankov, Philippe Gantzer, Dmitry Babadeev, Anna Pinigina, Timur Madzhidov, and Alexandre Varnek. Synplanner: An end-to-end tool for synthesis planning.Journal of Chemical Information and Modeling, 65(1):15–21, 2025. doi: 10.1021/acs.jcim.4c02004

work page doi:10.1021/acs.jcim.4c02004 2025

[32] [32]

Retrosynthesis planning via worst-path policy optimisation in tree-structured mdps, 2025

Mianchu Wang and Giovanni Montana. Retrosynthesis planning via worst-path policy optimisation in tree-structured mdps, 2025. URLhttps://arxiv.org/abs/2509.10504

arXiv 2025

[33] [33]

AI-driven synthetic route design incorporated with retrosynthesis knowledge.Journal of Chemical Information and Modeling, 62(6):1357–1367, 2022

Shoichi Ishida, Kei Terayama, Ryosuke Kojima, Kiyosei Takasu, and Yasushi Okuno. AI-driven synthetic route design incorporated with retrosynthesis knowledge.Journal of Chemical Information and Modeling, 62(6):1357–1367, 2022. doi: 10.1021/acs.jcim.1c01074

work page doi:10.1021/acs.jcim.1c01074 2022

[34] [34]

Comparing search algorithms on the retrosynthesis problem

Milo Roucairol and Tristan Cazenave. Comparing search algorithms on the retrosynthesis problem. Molecular Informatics, 43, 2024. URL https://api.semanticscholar.org/CorpusID:253882602

2024

[35] [35]

Multimodal large language models for inverse molecular design with retrosynthetic planning, 2024

Gang Liu, Michael Sun, Wojciech Matusik, Meng Jiang, and Jie Chen. Multimodal large language models for inverse molecular design with retrosynthetic planning, 2024. URL https://arxiv.org/abs/2410. 04223

2024

[36] [36]

Kevin Yu, Jihye Roh, Ziang Li, Wenhao Gao, Runzhong Wang, and Connor W. Coley. Double-ended synthesis planning with goal-constrained bidirectional search. InAdvances in Neural Information Pro- cessing Systems, volume 37, pages 112919–112949. Neural Information Processing Systems Foundation, Inc., 2024. doi: 10.52202/079017-3588. URL https://proceedings.ne...

work page doi:10.52202/079017-3588 2024

[37] [37]

Chemist-aligned retrosynthesis by ensembling diverse inductive bias models, 2025

Krzysztof Maziarz, Guoqing Liu, Hubert Misztela, Austin Tripp, Junren Li, Aleksei Kornev, Piotr Gai´nski, Holger Hoefling, Mike Fortunato, Rishi Gupta, and Marwin Segler. Chemist-aligned retrosynthesis by ensembling diverse inductive bias models, 2025. URLhttps://arxiv.org/abs/2412.05269

arXiv 2025

[38] [38]

Baker, Daniel Adu-Ampratwum, Reza Averly, Botao Yu, Huan Sun, and Xia Ning

Frazier N. Baker, Daniel Adu-Ampratwum, Reza Averly, Botao Yu, Huan Sun, and Xia Ning. LARC: Towards human-level constrained retrosynthesis planning through an agentic framework. InProceedings of AI for Accelerated Research Symposium, volume 3 ofEPiC Series in Technology, pages 153–176. EasyChair, 2026. doi: 10.29007/z3hb. URLhttps://easychair.org/publica...

work page doi:10.29007/z3hb 2026

[39] [39]

Bran, Zlatko Jonˇcev, and Philippe Schwaller

Nguyen Xuan-Vu, Daniel Armstrong, Milena Wehrbach, Andres M. Bran, Zlatko Jonˇcev, and Philippe Schwaller. Synthelite: Chemist-aligned and feasibility-aware synthesis planning with LLMs, 2025. URL https://arxiv.org/abs/2512.16424

arXiv 2025

[40] [40]

AOT*: Efficient synthesis planning via LLM-empowered AND-OR tree search, 2025

Xiaozhuang Song, Xuanhao Pan, Xinjian Zhao, Hangting Ye, Shufei Zhang, Jian Tang, and Tianshu Yu. AOT*: Efficient synthesis planning via LLM-empowered AND-OR tree search, 2025. URL https: //arxiv.org/abs/2509.20988

arXiv 2025

[41] [41]

Diverse and feasible retrosynthesis using GFlowNets.Information Sciences, 714:122194, 2025

Piotr Gai ´nski, Michał Koziarski, Krzysztof Maziarz, Marwin Segler, Jacek Tabor, and Marek ´Smieja. Diverse and feasible retrosynthesis using GFlowNets.Information Sciences, 714:122194, 2025. doi: 10.1016/j.ins.2025.122194. URLhttps://doi.org/10.1016/j.ins.2025.122194

work page doi:10.1016/j.ins.2025.122194 2025

[42] [42]

Nair, Rico Häuselmann, Riccardo Pisoni, Costas Bekas, Anna Iuliano, and Teodoro Laino

Philippe Schwaller, Riccardo Petraglia, Valerio Zullo, Vishnu H. Nair, Rico Häuselmann, Riccardo Pisoni, Costas Bekas, Anna Iuliano, and Teodoro Laino. Predicting retrosynthetic pathways using transformer- based models and a hyper-graph exploration strategy.Chemical Science, 11:3316 – 3325, 2020. URL https://api.semanticscholar.org/CorpusID:216332642

2020

[43] [43]

Automatic retrosynthetic route planning using template-free models.Chemical Science, 11:3355 – 3364, 2020

Kangjie Lin, Youjun Xu, Jianfeng Pei, and Luhua Lai. Automatic retrosynthetic route planning using template-free models.Chemical Science, 11:3355 – 3364, 2020. URL https://api.semanticscholar. org/CorpusID:268816571

2020

[44] [44]

Multistep retrosynthesis combining a disconnection aware triple transformer loop with a route penalty score guided tree search.Chemical Science, 14:9959 – 9969, 2023

David Kreutter and Jean-Louis Reymond. Multistep retrosynthesis combining a disconnection aware triple transformer loop with a route penalty score guided tree search.Chemical Science, 14:9959 – 9969, 2023. URLhttps://api.semanticscholar.org/CorpusID:261488982

2023

[45] [45]

Yu Shee, Anton Morgunov, Haote Li, and Victor S. Batista. DirectMultiStep: Direct route generation for multistep retrosynthesis.Journal of Chemical Information and Modeling, 65(8):3903–3914, 2025. doi: 10.1021/acs.jcim.4c01982

work page doi:10.1021/acs.jcim.4c01982 2025

[46] [46]

Fast and scalable retrosynthetic planning with a transformer neural network and speculative beam search

Natalia Andronova, Mikhail Andronov, Jürgen Schmidhuber, Michael Wand, and Djork-Arné Clevert. Fast and scalable retrosynthetic planning with a transformer neural network and speculative beam search. Digital Discovery, 5:1783–1793, 2026. doi: 10.1039/D5DD00573F. URL http://dx.doi.org/10. 1039/D5DD00573F

work page doi:10.1039/d5dd00573f 2026

[47] [47]

Cavanagh, Yingze Wang, Jacob M

Kunyang Sun, Dorian Bagni, Joseph M. Cavanagh, Yingze Wang, Jacob M. Sawyer, Bo Zhou, Andrew Gritsevskiy, Oufan Zhang, and Teresa Head-Gordon. Synllama: Generating synthesizable molecules and their analogs with large language models.ACS Central Science, 11(11):2108–2120, 2025. doi: 10.1021/acscentsci.5c01285

work page doi:10.1021/acscentsci.5c01285 2025

[48] [48]

TempRe: Template generation for single and direct multi-step retrosynthesis, 2025

Nguyen Xuan-Vu, Daniel P Armstrong, Zlatko Jon ˇcev, and Philippe Schwaller. TempRe: Template generation for single and direct multi-step retrosynthesis, 2025. URL https://arxiv.org/abs/2507. 21762

2025

[49] [49]

LLM-augmented chemical synthesis and design decision programs, 2025

Haorui Wang, Jeff Guo, Lingkai Kong, Rampi Ramprasad, Philippe Schwaller, Yuanqi Du, and Chao Zhang. LLM-augmented chemical synthesis and design decision programs, 2025. URL https://arxiv. org/abs/2505.07027

Pith/arXiv arXiv 2025

[50] [50]

Retrosynformer: planning multi-step chemical synthesis routes via a decision transformer.Digital Discovery, 5:348–362, 2026

Emma Granqvist, Rocío Mercado, and Samuel Genheden. Retrosynformer: planning multi-step chemical synthesis routes via a decision transformer.Digital Discovery, 5:348–362, 2026. doi: 10.1039/D5DD00153F. URLhttp://dx.doi.org/10.1039/D5DD00153F

work page doi:10.1039/d5dd00153f 2026

[51] [51]

PaRoutes: towards a framework for benchmarking retrosynthesis route predictions.Digital Discovery, 1:527–539, 2022

Samuel Genheden and Esben Bjerrum. PaRoutes: towards a framework for benchmarking retrosynthesis route predictions.Digital Discovery, 1:527–539, 2022. doi: 10.1039/D2DD00015F. URL http: //dx.doi.org/10.1039/D2DD00015F

work page doi:10.1039/d2dd00015f 2022

[52] [52]

Anton Morgunov and Victor S. Batista. Procrustean bed for AI-driven retrosynthesis: A unified framework for reproducible evaluation, 2025. URLhttps://arxiv.org/abs/2512.07079

arXiv 2025

[53] [53]

Choure, Mun Hong Fong, Jihye Roh, Itai Levin, Kevin Yu, Joonyoung F

Zhengkai Tu, Sourabh J. Choure, Mun Hong Fong, Jihye Roh, Itai Levin, Kevin Yu, Joonyoung F. Joung, Nathan Morgan, Shih-Cheng Li, Xiaoqi Sun, Huiqian Lin, Mark Murnin, Jordan P. Liles, Thomas J. Struble, Michael E. Fortunato, Mengjie Liu, William H. Green, Klavs F. Jensen, and Connor W. Coley. ASKCOS: Open-source, data-driven synthesis planning.Accounts o...

work page doi:10.1021/acs.accounts.5c00155 2025

[54] [54]

Unified language model pre-training for natural language understanding and generation

Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xiaodong Liu, Yu Wang, Jianfeng Gao, Ming Zhou, and Hsiao-Wuen Hon. Unified language model pre-training for natural language understanding and generation. InAdvances in Neural Information Processing Systems, volume 32, 2019. URL https://proceedings. neurips.cc/paper/2019/hash/c20bb2d9a50d5ac1f713f8b34d9aac5a-Abstr...

2019

[55] [55]

Richard S. Sutton. The bitter lesson, 2019. URL http://www.incompleteideas.net/IncIdeas/ BitterLesson.html

2019

[56] [56]

Attention is all you need

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In I. Guyon, U. V on Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc., 2017. URL htt...

2017

[57] [57]

doi: https://doi.org/10.1016/j.neucom

Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024. doi: 10.1016/j.neucom. 2023.127063. URLhttps://doi.org/10.1016/j.neucom.2023.127063

work page doi:10.1016/j.neucom 2024

[58] [58]

Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization, 2016. URL https: //arxiv.org/abs/1607.06450

Pith/arXiv arXiv 2016

[59] [59]

Root mean square layer normalization

Biao Zhang and Rico Sennrich. Root mean square layer normalization. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. URL https://proceedings.neurips.cc/paper_ files/paper/2019/file/1e8a19426224ca89e83cef47f1e7f53b-Paper.pdf

2019

[60] [60]

On layer normalization in the transformer architecture

Ruibin Xiong, Yunchang Yang, Di He, Kai Zheng, Shuxin Zheng, Chen Xing, Huishuai Zhang, Yanyan Lan, Liwei Wang, and Tie-Yan Liu. On layer normalization in the transformer architecture. InPro- ceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event, Proceedings of Machine Learning Research, pages 10524–1...

2020

[61] [61]

Gaussian error linear units (gelus), 2016

Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus), 2016. URL https://arxiv.org/ abs/1606.08415

Pith/arXiv arXiv 2016

[62] [62]

Glu variants improve transformer, 2020

Noam Shazeer. Glu variants improve transformer, 2020. URL https://arxiv.org/abs/2002.05202

Pith/arXiv arXiv 2020

[63] [63]

Llama: Open and efficient foundation language models, 2023

Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. Llama: Open and efficient foundation language models, 2023. URL https://arxiv.org/abs/2302.13971

Pith/arXiv arXiv 2023

[64] [64]

Decoupled weight decay regularization

Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net, 2019. URLhttps://openreview.net/forum?id=Bkg6RiCqY7

2019

[65] [65]

Muon: An optimizer for hidden layers in neural networks, 2024

Keller Jordan, Yuchen Jin, Vlado Boza, Jiacheng You, Franz Cesista, Laker Newhouse, and Jeremy Bern- stein. Muon: An optimizer for hidden layers in neural networks, 2024. URL https://kellerjordan. github.io/posts/muon/

2024

[66] [66]

Muon is scalable for LLM training, 2025

Jingyuan Liu, Jianlin Su, Xingcheng Yao, Zhejun Jiang, Guokun Lai, Yulun Du, Yidao Qin, Weixin Xu, Enzhe Lu, Junjie Yan, Yanru Chen, Huabin Zheng, Yibo Liu, Shaowei Liu, Bohong Yin, Weiran He, Han Zhu, Yuzhi Wang, Jianzhou Wang, Mengnan Dong, Zheng Zhang, Yongsheng Kang, Hao Zhang, Xinran Xu, Yutao Zhang, Yuxin Wu, Xinyu Zhou, and Zhilin Yang. Muon is sca...

Pith/arXiv arXiv 2025

[67] [67]

e3nn: Euclidean neural networks,

Reiner Pope, Sholto Douglas, Aakanksha Chowdhery, Jacob Devlin, James Bradbury, Anselm Levskaya, Jonathan Heek, Kefan Xiao, Shivani Agrawal, and Jeff Dean. Efficiently scaling transformer inference. CoRR, abs/2211.05102, 2022. doi: 10.48550/ARXIV .2211.05102. URL https://doi.org/10.48550/ arXiv.2211.05102. 13 Supporting Information The Supporting Informat...

work page internal anchor Pith review doi:10.48550/arxiv 2022

[68] [68]

Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered

Top-10 CI gives the 95% bootstrap interval. Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered. Paired CI rows compare Ariadne with AiZynthFinder MCTS: intervals containing zero indicate no significant difference, positive intervals favor Ariadne, and n...

[69] [69]

Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered

Top-10 CI gives the 95% bootstrap interval. Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered. Paired CI rows compare Ariadne with AiZynthFinder MCTS: intervals containing zero indicate no significant difference, positive intervals favor Ariadne, and n...

[70] [70]

Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered

Top-10 CI gives the 95% bootstrap interval. Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered. Paired CI rows compare Ariadne with AiZynthFinder MCTS: intervals containing zero indicate no significant difference, positive intervals favor Ariadne, and n...

[71] [71]

Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered

Top-10 CI gives the 95% bootstrap interval. Root T10 measures recovery of the reference root reaction, and Route|root T10 measures full-route recovery among targets whose root reaction was recovered. Paired CI rows compare Ariadne with AiZynthFinder MCTS: intervals containing zero indicate no significant difference, positive intervals favor Ariadne, and n...