arxiv: 2605.00313 · v1 · submitted 2026-05-01 · ❄️ cond-mat.mtrl-sci · cs.AI

Recognition: unknown

Beyond Structure: Revolutionising Materials Discovery via AI-Driven Synthesis Protocol-Property Relationships

Guillaume Lambard

Authors on Pith no claims yet

Pith reviewed 2026-05-09 19:43 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci cs.AI

keywords synthesis protocolsmaterials discoverysynthesizability gapgenerative modelsclosed-loop optimizationmachine-readable protocolsAI-driven discovery

0 comments

The pith

To close the synthesizability gap, AI materials discovery must treat executable synthesis protocols as primary design variables rather than atomic structures.

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

The paper argues that the prevailing structure-centric approach in AI-driven materials discovery generates many candidate atomic configurations but leaves most of them unrealizable in the lab, creating a persistent synthesizability gap. It proposes a pivot to a synthesis-first paradigm that elevates executable synthesis protocols to the role of main design variables. The argument rests on representing protocols in machine-readable form, applying generative models to suggest complete reaction pathways, and using closed-loop optimization to adjust protocols against real experimental outcomes and sustainability needs. A sympathetic reader would care because this reframing could convert theoretical AI outputs into materials that laboratories can actually produce and verify.

Core claim

By centering materials discovery on the causal backbone from synthesis protocol P to structure X to property y, and by constructing AI systems on the three pillars of machine-readable protocols, generative inverse-design models for reaction recipes, and closed-loop experimental refinement, the field can move beyond the current barrier where predicted structures rarely translate into synthesizable materials.

What carries the argument

The causal backbone P→X→y linking synthesis protocol to structure to properties, together with the three pillars of machine-readable protocol representations, generative models for proposing actionable pathways, and closed-loop optimization against experimental feedback.

If this is right

Generative models will output complete, executable synthesis recipes instead of isolated atomic arrangements.
Self-driving laboratory platforms can directly test and refine protocol predictions in iterative cycles.
Sustainability metrics and experimental feasibility become explicit optimization targets within the design process.
Experimental results feed back immediately to update and improve the underlying protocol-generation models.

Where Pith is reading between the lines

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

Laboratories could reduce wasted effort on unworkable syntheses by screening for feasible protocols at the design stage.
Standardized machine-readable protocol formats might enable broader sharing and reuse of synthesis data across independent groups.
The same protocol-centric logic could be tested in adjacent domains such as chemical process optimization or formulation design.

Load-bearing premise

That elevating synthesis protocols to primary design variables and implementing the three pillars will be enough to make AI-proposed materials routinely achievable in practice.

What would settle it

An experiment in which a system built on machine-readable protocols, generative recipe models, and closed-loop optimization produces no higher fraction of successfully synthesized materials than current structure-only AI methods.

Figures

Figures reproduced from arXiv: 2605.00313 by Guillaume Lambard.

**Figure 1.** Figure 1: Paradigm shift from structure-centric to synthesis protocol-centric discovery. (A) Conventional view at source ↗

**Figure 2.** Figure 2: Integrated ecosystem linking heterogeneous data sources, protocol representations, AI/ML core view at source ↗

read the original abstract

The current structure-centric paradigm in artificial intelligence (AI)-driven materials discovery, despite delivering thousands of candidate structures, is stalling at a critical barrier: the synthesizability gap. We argue that closing this gap demands a pivot to a synthesis-first paradigm in which executable synthesis protocols, not just atomic configurations, are treated as primary design variables. We outline a roadmap built on three pillars: (i) representing synthesis procedures as machine-readable protocols, (ii) deploying generative and inverse-design models to propose actionable reaction pathways and recipes, and (iii) integrating closed-loop optimisation to refine protocols against experimental realities and sustainability constraints. Framed in terms of the causal backbone P->X->y from protocol P to structure X and properties y, this perspective sets out methodological building blocks, standards needs and self-driving laboratory (SDL) integration strategies to accelerate reproducible, data-first materials discovery.

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

This perspective makes a clear case for treating synthesis protocols as primary design variables but leaves the data problem for generative models unaddressed.

read the letter

The core takeaway is that the paper advocates shifting materials discovery from structure-centric AI to a synthesis-first approach where executable protocols become the main variables. It frames this with a P to X to y causal chain and a three-pillar roadmap covering machine-readable protocols, generative models for pathways, and closed-loop optimization tied to experiments and sustainability. That framing is useful because it directly targets the synthesizability gap that has stalled many structure-prediction efforts so far. The authors pull together existing threads from self-driving lab work into a coherent perspective without overclaiming novelty for the individual pieces. This gives readers a practical way to think about integrating AI with real lab constraints. The paper does well at highlighting reproducibility needs and the value of including failure data in future datasets. It stays grounded in the current limitations of the field rather than promising quick fixes. The main soft spot is that pillar two, the generative models for proposing actionable recipes, will require large-scale paired data on protocols, conditions, and outcomes. The text calls for machine-readable standards but provides no mechanism or strategy for building that dataset at the necessary volume and diversity. Without addressing how to collect and standardize those records, including negatives, the inverse-design step remains aspirational. As a perspective piece there are no new derivations, examples, or tests to evaluate the pillars against real cases. This work is aimed at researchers in AI for materials and self-driving labs who are already thinking about workflow changes. It offers a high-level map rather than immediate tools or results, so it will appeal most to those planning longer-term lab or funding strategies. I would send it for peer review. The argument is coherent enough to benefit from community input on implementation details even if it stays at the perspective level.

Referee Report

2 major / 2 minor

Summary. The manuscript is a perspective piece arguing that the structure-centric paradigm in AI-driven materials discovery has reached a synthesizability barrier. It proposes pivoting to a synthesis-first paradigm that elevates executable synthesis protocols as primary design variables rather than atomic structures alone. The authors outline a three-pillar roadmap—(i) machine-readable synthesis protocols, (ii) generative and inverse-design models for reaction pathways, and (iii) closed-loop optimization integrated with self-driving laboratories—framed by the causal chain P→X→y from protocol to structure to property, together with calls for standards and SDL integration to enable reproducible discovery.

Significance. If the roadmap can be realized, the shift could meaningfully reduce the gap between computationally predicted materials and experimentally accessible ones, improving the practical utility of AI in materials science. The perspective earns credit for its explicit causal framing P→X→y, which organizes the argument cleanly, and for identifying concrete needs around machine-readable representations and SDL integration as actionable next steps.

major comments (2)

[Pillar (ii)] Pillar (ii) section: the discussion of generative and inverse-design models for proposing actionable synthesis pathways does not supply any mechanism, standard, or acquisition strategy for obtaining the large-scale paired datasets of protocols, conditions, and outcomes (including failures) that such models require for training. This is load-bearing for the central claim, as the feasibility of pillar (ii) rests on the existence of such data.
[Roadmap] Roadmap overview: the assertion that the three pillars together will close the synthesizability gap is advanced without any concrete examples, partial implementations, or preliminary evidence showing that machine-readable protocols can be generated at scale or that closed-loop systems can converge on viable recipes. This leaves the sufficiency of the proposed pillars untested within the manuscript.

minor comments (2)

[Introduction] The notation P→X→y is introduced in the abstract and used throughout but would benefit from an explicit mapping to experimental variables (e.g., temperature, solvent choice) in the introduction to avoid ambiguity for readers outside the immediate subfield.
References to existing machine-readable protocol efforts (e.g., in organic synthesis or materials databases) are mentioned at a high level; adding one or two specific citations with brief comparison would strengthen the standards-needs discussion.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and insightful comments on our perspective manuscript. We address each major comment point by point below, indicating where we will revise the text to incorporate the feedback while preserving the forward-looking nature of the piece.

read point-by-point responses

Referee: [Pillar (ii)] Pillar (ii) section: the discussion of generative and inverse-design models for proposing actionable synthesis pathways does not supply any mechanism, standard, or acquisition strategy for obtaining the large-scale paired datasets of protocols, conditions, and outcomes (including failures) that such models require for training. This is load-bearing for the central claim, as the feasibility of pillar (ii) rests on the existence of such data.

Authors: We agree that the availability of large-scale, high-quality paired datasets (including failures) is essential for training the generative and inverse-design models discussed in Pillar (ii). The original manuscript identifies the need for such models but does not detail acquisition mechanisms, as its primary aim is to articulate the broader paradigm shift. In the revised version we will expand this section to outline concrete strategies, including automated literature mining via NLP and information extraction tools to parse protocols and outcomes from existing publications, community-driven standards for depositing machine-readable recipes, and the systematic generation of new paired data through high-throughput experimentation in self-driving laboratories. These additions will strengthen the discussion of feasibility without changing the perspective character of the work. revision: yes
Referee: [Roadmap] Roadmap overview: the assertion that the three pillars together will close the synthesizability gap is advanced without any concrete examples, partial implementations, or preliminary evidence showing that machine-readable protocols can be generated at scale or that closed-loop systems can converge on viable recipes. This leaves the sufficiency of the proposed pillars untested within the manuscript.

Authors: The referee correctly observes that the manuscript presents the three-pillar roadmap as a conceptual solution without including specific case studies or quantitative demonstrations of scalability. Because this is a perspective proposing a new research direction rather than a report of completed implementations, such evidence was not originally included. We will revise the Roadmap section to reference emerging partial implementations already appearing in the literature, such as NLP-based extraction of synthesis protocols at scale in selected chemical domains and early closed-loop SDL demonstrations that have refined recipes for specific materials. These citations will illustrate current progress on individual pillars while clarifying that their full integration to close the synthesizability gap remains prospective, thereby moderating the claim to one of promising potential grounded in ongoing developments. revision: yes

Circularity Check

0 steps flagged

No circularity: perspective roadmap without derivations or self-referential reductions

full rationale

This is a forward-looking perspective paper that argues for a synthesis-first paradigm and outlines three conceptual pillars (machine-readable protocols, generative models, closed-loop optimization) framed around a causal P→X→y backbone. It contains no equations, no fitted parameters, no predictions derived from data, and no mathematical derivations. The central claims are argumentative and programmatic rather than obtained by reducing any quantity to its own inputs or to a self-citation chain. No load-bearing uniqueness theorems, ansatzes, or renamings of known results appear. Self-citations, if present, are not invoked to justify the core roadmap as an external fact. The paper is therefore self-contained as a viewpoint and exhibits no circularity by the defined criteria.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper introduces no free parameters or invented entities. It rests on domain assumptions about the current state of AI materials discovery and the efficacy of the proposed shift. Because only the abstract is available, the ledger is necessarily incomplete.

axioms (2)

domain assumption The structure-centric paradigm in AI materials discovery is stalling at the synthesizability gap
Presented as the critical barrier that motivates the entire perspective.
ad hoc to paper Executable synthesis protocols can be represented as machine-readable objects suitable for generative AI and closed-loop optimization
This is the foundational premise of the three-pillar roadmap.

pith-pipeline@v0.9.0 · 5443 in / 1411 out tokens · 41136 ms · 2026-05-09T19:43:58.516687+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

61 extracted references · 3 canonical work pages

[1]

Persson and Gerbrand Ceder

Kristin A. Persson and Gerbrand Ceder. A materials genome approach to accelerating materials inno- vation.MRS Bull., 38(9):743–750, 2013

2013
[2]

Materials genome initiative for global competitiveness

National Science and Technology Council. Materials genome initiative for global competitiveness. U.S. Executive Office of the President report, 2011. Available athttps://obamawhitehouse.archives.gov/ sites/default/files/microsites/ostp/materials_genome_initiative-final.pdf

2011
[3]

Richards, Stephen Dacek, Shreyas Cholia, Dan Gunter, David Skinner, Gerbrand Ceder, and Kristin A

Anubhav Jain, Shyue Ping Ong, Geoffroy Hautier, Wei Chen, William D. Richards, Stephen Dacek, Shreyas Cholia, Dan Gunter, David Skinner, Gerbrand Ceder, and Kristin A. Persson. Commentary: The materials project: A materials genome approach to accelerating materials innovation.APL Mater., 1(1):011002, 2013

2013
[4]

Taylor, Lance J

Stefano Curtarolo, Wahyu Setyawan, Xuejie Wang, Jianlin Xue, Kesong Yang, Richard H. Taylor, Lance J. Nelson, Gus L. W. Hart, Stefano Sanvito, Marco Buongiorno Nardelli, and Natalio Mingo. Aflow: An automatic framework for high-throughput materials discovery.Comput. Mater. Sci., 58:218– 226, 2012

2012
[5]

Saal, Scott Kirklin, Muratahan Aykol, Bryce Meredig, and Christopher Wolverton

James E. Saal, Scott Kirklin, Muratahan Aykol, Bryce Meredig, and Christopher Wolverton. Materials design and discovery with high-throughput density functional theory: The open quantum materials database (oqmd).JOM, 65(11):1501–1509, 2013. 16

2013
[6]

Beware of plausible predictions of fantasy materials.Nature, 566:447–449, 2019

Alex Zunger. Beware of plausible predictions of fantasy materials.Nature, 566:447–449, 2019. Discusses the low ( 2%) experimental realisation rate of >50,000 hypothetical low-enthalpy phases proposed by high-throughput DFT surveys

2019
[7]

Coley, William H

Connor W. Coley, William H. Green, and Klavs F. Jensen. Machine learning in computer-aided synthesis planning.Acc. Chem. Res., 51(5):1281–1289, 2018

2018
[8]

Coley, Luke Rogers, William H

Connor W. Coley, Luke Rogers, William H. Green, and Klavs F. Jensen. Computer-assisted retrosyn- thesis based on molecular similarity.ACS Cent. Sci., 3(12):1237–1245, 2017

2017
[9]

Coley, Natalie S

Connor W. Coley, Natalie S. Eyke, and Klavs F. Jensen. Autonomous discovery in the chemical sciences: Part i – progress.Nat. Rev. Mater., 5:766–779, 2020

2020
[10]

A self-driving lab for nano- and advanced ma- terials synthesis.ACS Nano, 19(9):9029–9041, 2025

Mohammad Zaki, Carsten Prinz, and Bastian Ruehle. A self-driving lab for nano- and advanced ma- terials synthesis.ACS Nano, 19(9):9029–9041, 2025. Describes the MINERVA platform for automated synthesis, purification, and in-line characterisation of nano- and advanced materials

2025
[11]

Aizynthfinder: a fast, robust and flexible open- source software for retrosynthetic planning.J

Samuel Genheden, Ajay Thakkar, and Ola Engkvist. Aizynthfinder: a fast, robust and flexible open- source software for retrosynthetic planning.J. Cheminform., 12:70, 2020

2020
[12]

Inverse molecular design using machine learning: Generative models for matter engineering.Science, 361(6400):360–365, 2018

Benjamin Sanchez-Lengeling and Alán Aspuru-Guzik. Inverse molecular design using machine learning: Generative models for matter engineering.Science, 361(6400):360–365, 2018

2018
[13]

Wei, David Duvenaud, José Miguel Hernández-Lobato, Benjamin Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D

Rafael Gómez-Bombarelli, Jennifer N. Wei, David Duvenaud, José Miguel Hernández-Lobato, Benjamin Sánchez-Lengeling, Dennis Sheberla, Jorge Aguilera-Iparraguirre, Timothy D. Hirzel, Ryan P. Adams, and Alán Aspuru-Guzik. Automatic chemical design using a data-driven continuous representation of molecules.ACS Cent. Sci., 4(2):268–276, 2018

2018
[14]

drugan: An advanced generative adversarial autoencoder model for de novo generation of molecules.Molecular Pharmaceutics, 14(9):3098–3104, 2017

Artur Kadurin, Sergey Nikolenko, Alex Aliper, and et al. drugan: An advanced generative adversarial autoencoder model for de novo generation of molecules.Molecular Pharmaceutics, 14(9):3098–3104, 2017

2017
[15]

Equivariant diffusion for molecule generation in 3d

Emiel Hoogeboom and colleagues. Equivariant diffusion for molecule generation in 3d. InProceedings of NeurIPS, 2022

2022
[16]

E(n) equivariant graph neural networks

Victor Garcia Satorras, Emiel Hoogeboom, and Max Welling. E(n) equivariant graph neural networks. InProceedings of the 38th International Conference on Machine Learning, 2021

2021
[17]

Tabor, Laurent M

Daniel P. Tabor, Laurent M. Roch, Semion K. Saikin, Christoph Kreisbeck, Dennis Sheberla, Joseph H. Montoya, Shyam Dwaraknath, Muratahan Aykol, Christopher Ortiz, Henning Tribukait, Carlos Amador-Bedolla, Christoph J. Brabec, Vladimir Bulovic, Moungi G. Bawendi, and Alán Aspuru-Guzik. Accelerating the discovery of materials for clean energy in the era of ...

2018
[18]

Wei Gao and Connor W. Coley. The synthesizability of molecules proposed by generative models.J. Am. Chem. Soc., 144:14813–14824, 2022

2022
[19]

Estimation of synthetic accessibility of drug-like molecules from molecular complexity and fragment contributions.J

Peter Ertl and Ansgar Schuffenhauer. Estimation of synthetic accessibility of drug-like molecules from molecular complexity and fragment contributions.J. Cheminform., 1(1):8, 2009

2009
[20]

Merkel and Alán Aspuru-Guzik

Daniel J. Merkel and Alán Aspuru-Guzik. Kinetic traps and how to avoid them: Integrating reaction- path information into materials discovery.J. Am. Chem. Soc., 143(45):18821–18832, 2021

2021
[21]

Environmental Protection Agency

U.S. Environmental Protection Agency. Toxic substances control act (tsca) chemical substance inven- tory.https://www.epa.gov/tsca-inventory, 2023. Accessed 15 June 2024

2023
[22]

Kovács, Gregor N

Ilyes Batatia, Dávid P. Kovács, Gregor N. C. Simm, Christoph Ortner, and Gábor Csányi. Mace: Higher-order equivariant message passing neural networks for fast and accurate molecular force fields. arXiv, 2022. 17

2022
[23]

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.J. Chem. Inf. Model., 58(2):252–261, 2018

2018
[24]

Bjerrum, Ola Engkvist, and Jean-Louis Reymond

Ajay Thakkar, Veronika Chadimová, Emil J. Bjerrum, Ola Engkvist, and Jean-Louis Reymond. Ret- rosynthetic accessibility score (RAscore)—rapid machine-learned synthesizability classification from ai- driven retrosynthetic planning.Chem. Sci., 12:3339–3349, 2021

2021
[25]

Predicting synthesizability of crystalline materials via deep learning.Commun

Ali Davariashtiyani, Zahra Kadkhodaie, and Sara Kadkhodaei. Predicting synthesizability of crystalline materials via deep learning.Commun. Mater., 2:115, 2021

2021
[26]

Sequential translation normalizing flows for molecular graph generation

Max Jacques et al. Sequential translation normalizing flows for molecular graph generation. InProceed- ings of ICML, 2021

2021
[27]

Granda, Gareth Keenan, and et al

Schel Roland Steiner, Jonathan Wolf, Stefan Glatzel, Andreas Andreou, Jose M. Granda, Gareth Keenan, and et al. Organic synthesis in a modular robotic system driven by a chemical programming language.Science, 363(6423):eaav2211, 2019

2019
[28]

Schweidtmann, Andrew D

Artur M. Schweidtmann, Andrew D. Clayton, Luis Schenck, and et al. Machine learning meets contin- uous flow chemistry: Automated optimization toward the pareto front of multiple objectives.Chem. Eng. J., 417:129481, 2021

2021
[29]

Engel and Patrick A

Emily A. Engel and Patrick A. Sims. Non-equilibrium pathways govern polymorph selection during crystallisation.Nat. Mater., 20:178–185, 2021

2021
[30]

James Sherwood and James H. Clark. Green chemistry metrics and indicators.ACS Sustainable Chem. Eng., 9:7587–7602, 2021

2021
[31]

Autoprotocol: A standard for description of biological workflows

Transcriptic. Autoprotocol: A standard for description of biological workflows. https://autoprotocol.org/, 2018. Accessed 2025-08-09

2018
[32]

Paml: Protocol activity modeling language

Chris Myers, Anil Wipat, et al. Paml: Protocol activity modeling language. https://paml.org/, 2021. Accessed 2025-08-09

2021
[33]

Sila 2 – standardization in laboratory automation

SiLA Consortium. Sila 2 – standardization in laboratory automation. https://sila-standard.com/, 2020. Accessed 2025-08-09

2020
[34]

Animl – analytical information markup language

ASTM E13.15 Subcommittee. Animl – analytical information markup language. https://animl.org/,
[35]

Hastings et al

J. Hastings et al. The reaction ontology (rxno): Capturing chemical reactions in an ontological frame- work.J. Cheminform., 9:34, 2017

2017
[36]

Chebi in 2016: Improved services and an expanding collection of metabolites

Janna Hastings et al. Chebi in 2016: Improved services and an expanding collection of metabolites. Nucleic Acids Res., 44(D1):D1214–D1219, 2016

2016
[37]

MacLeod, Fraser G

Benjamin P. MacLeod, Fraser G. L. Parlane, Thomas D. Morrissey, Florian H"ase, Lo"ic M. Roch, and et al. Self-driving laboratory for accelerated discovery of thin-film materials.Sci. Adv., 6(20):eaaz8867, 2020

2020
[38]

Karniadakis

Maziar Raissi, Paris Perdikaris, and George E. Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving PDEs.J. Comput. Phys., 378:686–707, 2019

2019
[39]

Model-agnostic meta-learning for fast adaptation of deep networks

Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks. InProceedings of the 34th International Conference on Machine Learning, 2017

2017
[40]

Meta-learning for Bayesian optimization of black-box functions

Sarra Belakaria, Aviral Deshwal, Robisson Srinivasan, and Jayant Kalagnanam. Meta-learning for Bayesian optimization of black-box functions. InProceedings of the 38th International Conference on Machine Learning, 2021. 18

2021
[41]

Hyperband: A Novel Bandit-Based Approach to Hyperparameter Optimization

Lisha Li, Kevin Jamieson, Giulia DeSalvo, Afshin Rostamizadeh, and Ameet Talwalkar. Hyperband: A novel bandit-based approach to hyperparameter optimization. InProceedings of the International Conference on Learning Representations (ICLR), 2017. arXiv:1603.06560

work page Pith review arXiv 2017
[42]

Bohb: Robust and efficient hyperparameter optimiza- tion at scale

Stefan Falkner, Aaron Klein, and Frank Hutter. Bohb: Robust and efficient hyperparameter optimiza- tion at scale. InProceedings of the 35th International Conference on Machine Learning (ICML), pages 1437–1446, 2018

2018
[43]

Constrained bayesian opti- mization with noisy experiments.arXiv, 2018

Benjamin Letham, Brian Karrer, Guilherme Ottoni, and Eytan Bakshy. Constrained bayesian opti- mization with noisy experiments.arXiv, 2018

2018
[44]

Zhenpeng Zhou, Xiaocheng Li, and Richard N. Zare. Optimizing chemical reactions with deep rein- forcement learning.ACS Cent. Sci., 3(12):1337–1344, 2017

2017
[45]

Arnold, R

Lynton Ardizzone, Jakob Kruse, Sebastian Wirkert, Daniel Rahner, Eric W. Pellegrini, Ralf S. Klessen, Lena Maier-Hein, Carsten Rother, and Ullrich Köthe. Analyzing inverse problems with invertible neural networks.arXiv, 2018. arXiv:1808.04730

work page arXiv 2018
[46]

Panteleev, and et al

Andrea Bucci, Kevin Tran, Pavel A. Panteleev, and et al. On-the-fly closed-loop materials discovery via bayesian active learning.Nat. Commun., 11:250, 2020

2020
[47]

Roch, Dieter Kreutter, Matteo Aldeghi, and Alán Aspuru-Guzik

Florian Hase, Lo"ic M. Roch, Dieter Kreutter, Matteo Aldeghi, and Alán Aspuru-Guzik. Phoenics: A bayesian optimizer for chemistry.ACS Central Science, 4(10):1134–1145, 2018. Includes discussion of job-shop scheduling and failure recovery in autonomous laboratories

2018
[48]

Surya Prakash Rao, and U

Arihanth Srikar Tadanki, H. Surya Prakash Rao, and U. Deva Priyakumar. Dissecting errors in machine learning for retrosynthesis: A granular metric framework and a transformer-based model for more informative predictions.Digit. Discovery, 4:831–845, 2025. Introduces the SynFormer Transformer for reaction-sequence generation

2025
[49]

Epps, Fazel Bateni, Suyong Han, Kristofer G

Kameel Abdel-Latif, Robert W. Epps, Fazel Bateni, Suyong Han, Kristofer G. Reyes, and Milad Abol- hasani. Self-driven multistep quantum dot synthesis enabled by autonomous robotic experimentation in flow.Adv. Intell. Syst., 3(2):2000245, 2021

2021
[50]

Attia, Aditya Grover, Noel Jin, and et al

Peter M. Attia, Aditya Grover, Noel Jin, and et al. Closed-loop optimization of fast-charging protocols for batteries with machine learning.Nature, 578:397–402, 2020

2020
[51]

Murakami

Naoto Eguchi, Taro Fukazawa, Hiroyuki Kanda, Kohei Yamamoto, Takashi Miyake, and Takurou N. Murakami. Performance optimization of perovskite solar cells with an automated spin coating system and artificial intelligence technologies.EES Sol., 1:320–330, 2025

2025
[52]

Machine-learning- assisted and real-time-feedback-controlled growth of InAs/GaAs quantum dots.Nat

Chao Shen, Wenkang Zhan, Kaiyao Xin, Manyang Li, Zhenyu Sun, Hui Cong, Chi Xu, Jian Tang, Zhaofeng Wu, Bo Xu, Zhongming Wei, Chunlai Xue, Chao Zhao, and Zhanguo Wang. Machine-learning- assisted and real-time-feedback-controlled growth of InAs/GaAs quantum dots.Nat. Commun., 15:2724, 2024

2024
[53]

Kevin Tran and Zachary W. Ulissi. Active learning across intermetallics to guide discovery of electro- catalysts for co2 reduction and hydrogen evolution.Nat. Catal., 3:647–655, 2020

2020
[54]

Lundberg and Su-In Lee

Scott M. Lundberg and Su-In Lee. A unified approach to interpreting model predictions. InAdvances in Neural Information Processing Systems, pages 4765–4774, 2017

2017
[55]

Machine-learning-assisted discovery of rare-earth- free hard magnets.Adv

Daniel Scheiber, Manuel Richter, and Peter Kratzer. Machine-learning-assisted discovery of rare-earth- free hard magnets.Adv. Funct. Mater., 32:2109382, 2022

2022
[56]

Butler, Daniel W

Keith T. Butler, Daniel W. Davies, Hugh Cartwright, Olexandr Isayev, and Aron Walsh. Machine learning for molecular and materials science.Nature, 559:547–555, 2018

2018
[57]

Cooper, and Matthew R

Rebekah Jones, Ashley I. Cooper, and Matthew R. Hartings. Autonomous pet–raft polymerization platform for rapid library generation.Adv. Funct. Mater., 32:2107777, 2022. 19

2022
[58]

Elbert, Philip D

Paul Raccuglia, Katherine C. Elbert, Philip D. F. Adler, Casey Falk, Malia B. Wenny, Aurelio Mollo, Sorelle A. Friedler, Joshua Schrier, and Alexander J. Norquist. Machine-learning-assisted materials discovery using failed experiments.Nature, 533:73–76, 2016

2016
[59]

Statt, Brian A

Michael J. Statt, Brian A. Rohr, Kris Brown, Dan Guevarra, Jens Hummelshøj, Linda Hung, Abraham Anapolsky, John M. Gregoire, and Santosh K. Suram. Esamp: event-sourced architecture for materials provenance management and application to accelerated materials discovery.Digit. Discovery, 2:1078– 1088, 2023

2023
[60]

Elias, Ryan Chard, Maksim Levental, Zhengchun Liu, Ian Foster, and Santanu Chaudhuri

Jakob R. Elias, Ryan Chard, Maksim Levental, Zhengchun Liu, Ian Foster, and Santanu Chaudhuri. Real-time streaming and event-driven control of scientific experiments.arXiv, 2022. arXiv:2205.01476

work page arXiv 2022
[61]

Vaucher, Federico Zipoli, Jasper Geluykens, Vishnu H

Alain C. Vaucher, Federico Zipoli, Jasper Geluykens, Vishnu H. Nair, Vivek K. Parvathaneni, and Connor W. Coley. The open reaction database: Infrastructure for reaction informatics and machine learning.J. Chem. Inf. Model., 62:6475–6482, 2022. 20

2022