The DOI matches its Crossref record; 53 of the 53 references it cites were found in a registry or an index; 99 of the 163 sources its claims cite were checked for support.
From Scientific Copilots to Tool-Grounded Autonomy: AI Agents in Simulation-Driven Materials Discovery
Open accessAbstract
Artificial intelligence (AI) agents and large language model (LLM) agents are beginning to move materials discovery beyond isolated prediction tasks and toward tool-grounded workflows that can retrieve prior knowledge, configure simulations, launch calculations, inspect outputs, and decide what to do next. However, adjacent reviews on materials informatics, self-driving laboratories, natural-language processing in materials science, and autonomous chemistry have not isolated simulation-driven materials workflows as a distinct evidence base. This review addresses that gap through PRISMA-guided searches in Scopus (8 May 2026) and Web of Science (15 June 2026) for English-language journal articles published between 2022 and 2026. The combined search returned 232 records; 27 full texts were assessed and 26 studies were included in the final qualitative synthesis after one full-text exclusion. No eligible study was published in 2022 or 2023, indicating that the field emerged only in 2024 and expanded rapidly in 2025–2026. Catalysis and adsorption tasks (n = 6) and alloy design or evaluation (n = 5) dominated the corpus, while specialized multi-agent architectures were the most common pattern (n = 13). Across the included studies, agentic reasoning was most often coupled to workflow orchestration or integration tools, molecular-dynamics or atomistic simulation environments, and materials-data or machine learning screening pipelines; public repositories or archival artifacts were reported in 18 of 26 studies, experimental validation in six, and robotic closed-loop execution in only one study. The strongest evidence came from workflows that grounded language model decisions in simulators, structured databases, or experimentally verifiable outputs rather than in free-form text alone. This review therefore establishes AI agent workflow orchestration as a distinct analytical category within materials discovery and identifies the reporting, validation, and reproducibility conditions required for these systems to function as credible scientific infrastructure rather than as conversational demonstrations.
Overview
CoChat Summary
TL;DR
This PRISMA-guided systematic review synthesizes 26 studies on AI-agent workflows for simulation-driven materials discovery, finding a field that emerged in 2024 and expanded rapidly through 2025–2026. The strongest evidence comes from tool-grounded orchestration—agents coupled to simulators, databases, and analysis pipelines—rather than free-form language reasoning alone.
Key findings
- The final corpus comprised 26 studies from 232 combined records; no eligible study appeared in 2022 or 2023, with 1 in 2024, 9 in 2025, and 16 in the partial 2026 segment. Catalysis/adsorption (n = 6) and alloy design/evaluation (n = 5) dominated, and specialized multi-agent architectures were the most common pattern (n = 13).
- Tool grounding was broad but uneven: workflow orchestration/integration appeared in 21 of 26 studies, materials-data or ML screening pipelines in 11, molecular-dynamics or atomistic simulation environments in 10, explicit DFT/electronic-structure tools in 7, porous-material curation in 2, and robotic execution in 1.
- Quantitative highlights included Adsorb-Agent recovering comparable adsorption energies in 83.7% of benchmark systems and better energies in 35.0% while using only 26.9% of candidate configurations on average; a crystal-generation framework reducing execution-level hallucinations from 100% to as low as 5% and reaching 82% structural match for fully specified prompts; and an alloy-optimization loop improving tensile-strength estimates from 8.26–9.81 GPa to maxima of 13.31–18.70 GPa.
- Reproducibility and validation remained uneven: public repositories or archival artifacts were present in 18 of 26 studies, experimental validation in 6, high-autonomy workflows in 6, and genuine robotic closed-loop execution in only 1.
Methods
- Systematic review reported in accordance with PRISMA 2020, with searches in Scopus (8 May 2026) and Web of Science Core Collection (15 June 2026) for English-language journal articles published 2022–2026.
- Eligibility required an AI agent, LLM agent, autonomous agent, or multi-agent workflow as an operational part of the study; a materials discovery/design/optimization or closely related computational materials task; and a materially relevant simulation, computational modeling, screening, code execution, or domain tool component. Reviews, perspectives, editorials, conference papers, book chapters, and notes were excluded.
- Screening reduced 232 records to 76 unique English-language articles, then to 27 full texts and 26 included studies after one full-text exclusion. Data extraction was performed by one researcher with the other approving the final corpus; synthesis was comparative and qualitative because the evidence base was methodologically heterogeneous and a pooled meta-analysis was not pursued.
Limitations
- The paper states that the review was limited to Scopus and Web of Science Core Collection, restricted to English-language journal articles, and constrained by unstable agentic AI terminology; the 2026 segment is partial because searches were executed on 8 May 2026 and 15 June 2026. It also notes that all reported proportions derive from n = 26, so each inclusion or exclusion shifts a proportion by roughly 3.8 percentage points, and that studies indexed only in other repositories or released as preprints may be under-represented.
- The paper states that the eligibility criteria requiring a simulation, computational, or domain-specific tool component produced a corpus composed solely of tool-based studies, creating a selection effect; the importance of tool grounding should not be interpreted as evidence that it is more prevalent than non-tool-based methods in the general literature.
- The text makes evident additional constraints: data extraction was performed by only one researcher, the retained studies differ substantially in material domain, toolchain, and evidence maturity, and many papers lacked complete prompt logs, failure traces, scheduler settings, or environment specifications, making fair comparison and distinguishing genuine autonomy from curated demonstrations difficult.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
25 Valid · 9 Not valid · 7 Not checked · 89 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In parallel, chemistry-oriented reviews have shown that large language models can move beyond text generation toward tool use, retrieval, planning, code execution, and multi-agent coordination [5-10].”
What the sources say
- Supportedcites [5] Ramos, 2025, A review of large language models and autonomous agents in chemistry, Chem. Sci. source
The review’s “LLM-based autonomous agents” section describes tool use and execution of external code; its “Planning and reasoning modules” section discusses retrieval-augmented generation and planning. The “Agents for chemical innovation” section describes a multi-agent system whose agents collaborate on protein-design tasks. No page numbers are available in our copy. (LLM, conf=0.92)
- Partly supportedcites [6] Yoshikawa, 2023, Large language models for chemistry robotics, Auton. Robot. sourceExperimental — accuracy not yet measured
Section 2.2 discusses LLMs using external chemistry tools and this paper's approach to low-level planning; the Discussion describes generating plans that can be executed by a chemistry robot. The paper is an original robotics study, not a chemistry-oriented review, and the passages do not establish retrieval, LLM code execution, or multi-agent coordination. (LLM, conf=0.86)
- Partly supportedcites [7] Bran, 2024, Augmenting large language models with chemistry tools, Nat. Mach. Intell. sourceExperimental — accuracy not yet measured
The Main section describes ChemCrow using tools and planning successive steps; Methods > Tools describes literature retrieval and a Python tool that runs code. But this paper presents a ChemCrow study, not a review, and the cited passages do not show multi-agent coordination. (LLM, conf=0.88)
- Partly supportedcites [8] Boiko, 2023, Autonomous chemical research with large language models, Nature sourceExperimental — accuracy not yet measured
The Main and Coscientist system architecture sections describe a chemistry-focused LLM system that uses tools for search and retrieval, planning, code execution, and coordination among modules, including multiple LLMs. The Discussion calls the paper a proof of concept, however, so describing this source as a review is inaccurate. Our copy provides no page numbers for these passages. (LLM, conf=0.86)
- Partly supportedcites [9] Jablonka, 2024, Leveraging large language models for predictive chemistry, Nat. Mach. Intell. sourceExperimental — accuracy not yet measured
The opening section says LLMs can write code, and the Approach section reports chemistry tasks beyond text generation, including classification, regression and inverse design. But this paper is an original predictive-chemistry study, not a review, and the passages do not establish tool use, retrieval, planning, code execution or multi-agent coordination. Writing code is not the same as executing it. (LLM, conf=0.86)
- Partly supportedcites [10] Jablonka, 2023, 14 examples of how LLMs can transform materials science and chemistry: A reflection on a large language model hackathon, Digit. Discov. sourceExperimental — accuracy not yet measured
Section 1.2, “Automation and novel interfaces” (page not provided), describes LLMs using external tools and search and planning a synthesis; section 1.3, “Knowledge extraction” (page not provided), describes retrieval from scientific literature. The figure captions describe LLM-generated code producing a visualization, which partly supports code execution. The passages do not establish multi-agent coordination: an LLM acting as an agent is not the same as multiple agents coordinating. (LLM, conf=0.86)
- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“To the best of our knowledge, no prior review has isolated simulation-driven, toolorchestrating AI agent workflows for materials discovery as a distinct evidence base with study-level extraction of autonomy, validation, and reproducibility [3-14].”
What the sources say
- Could not verifycites [3] Tom, 2024, Self-Driving Laboratories for Chemistry and Materials Science, Chem. Rev. sourceExperimental — accuracy not yet measured
The Introduction (Brief History; page unavailable) says this review focuses on self-driving laboratories, and Infrastructure > Software (page unavailable) discusses computational simulation workflows. Neither passage assesses whether earlier reviews isolated simulation-driven, tool-orchestrating AI agents for materials discovery or extracted autonomy, validation, and reproducibility at study level. Searches for those terms and for reviews of autonomous computational materials discovery found no such assessment in our copy. (LLM, conf=0.71)
- Not checked — no full text of the sourcecites [4] Cheng, 2026, Artificial intelligence-driven approaches for materials design and discovery, Nat. Mater. source
- Supportedcites [5] Ramos, 2025, A review of large language models and autonomous agents in chemistry, Chem. Sci. source
The Introduction (page unavailable) defines this paper’s scope as LLMs and autonomous agents in chemistry. Its ‘Agents for chemical innovation’ section discusses materials datasets and atomistic simulation tools, but the review does not present simulation-driven materials-discovery workflows as a distinct evidence base with study-level extraction of autonomy, validation, and reproducibility. It supports the citation as an example of a broader prior review, not as proof of the citing authors’ field-wide novelty assessment. (LLM, conf=0.84)
- Could not verifycites [6] Yoshikawa, 2023, Large language models for chemistry robotics, Auton. Robot. sourceExperimental — accuracy not yet measured
Section 2.1, “Lab automation” (page not provided), discusses self-driving labs and cites a review of them. The Introduction (page not provided) describes this paper’s focus on chemistry robotics. Neither passage, nor the additional passages searched for simulation-driven workflows and study-level extraction, assesses whether a prior review covered the specific evidence base claimed. (LLM, conf=0.78)
- Could not verifycites [7] Bran, 2024, Augmenting large language models with chemistry tools, Nat. Mach. Intell. sourceExperimental — accuracy not yet measured
The Results and discussion section describes ChemCrow’s tool-orchestrated chemistry experiments, and the Conclusion discusses limits on reproducibility. Neither section establishes whether a prior review isolated simulation-driven agent workflows for materials discovery or extracted study-level evidence on autonomy, validation, and reproducibility. Searches for those review and evidence-extraction terms found no passage addressing the claim. (LLM, conf=0.73)
- Partly supportedcites [8] Boiko, 2023, Autonomous chemical research with large language models, Nature sourceExperimental — accuracy not yet measured
The Main section describes Coscientist as an autonomous, tool-using agent for chemical experiments, so this paper can serve as an example of a tool-orchestrating workflow. It is a primary experimental study, however, not a review of simulation-driven materials discovery; neither the Main nor Discussion establishes the citing authors’ claim that no prior review performed the specified study-level extraction. Page numbers were not provided in our copy. (LLM, conf=0.83)
- Could not verifycites [9] Jablonka, 2024, Leveraging large language models for predictive chemistry, Nat. Mach. Intell. sourceExperimental — accuracy not yet measured
The Classification and Concluding remarks sections discuss using language models for predictive chemistry, not the scope of prior reviews. Searches for simulation-driven, tool-orchestrating agents and for study-level extraction of autonomy, validation, and reproducibility found no passage assessing whether such a review existed. This paper therefore cannot verify the cited novelty claim. (LLM, conf=0.76)
- Could not verifycites [10] Jablonka, 2023, 14 examples of how LLMs can transform materials science and chemistry: A reflection on a large language model hackathon, Digit. Discov. sourceExperimental — accuracy not yet measured
Section 1.2 discusses LLM agents using external tools, and the Conclusion calls for better benchmarking and reproducibility. Neither passage addresses whether a prior review isolated simulation-driven agent workflows or extracted study-level evidence on autonomy, validation, and reproducibility; searches for those review criteria found no such passage. (LLM, conf=0.72)
- Partly supportedcites [11] Jiang, 2025, Applications of natural language processing and large language models in materials discovery, npj Comput. Mater. sourceExperimental — accuracy not yet measured
The Abstract describes a broad review of NLP and LLM applications in materials science, including autonomous research. The Introduction discusses tool-using agents and an example involving physics-based simulations. This supports citing the paper as a relevant prior review, but it does not establish that no prior review isolated simulation-driven agent workflows or extracted autonomy, validation, and reproducibility at the study level. (LLM, conf=0.78)
- Partly supportedcites [12] Pei, 2025, Language models for materials discovery and sustainability: Progress, challenges, and opportunities, Prog. Mater. Sci. sourceExperimental — accuracy not yet measured
Section 7.4 says the review mainly discusses experimental agents and that the calculation step needed for fully automated simulation workflows is currently missing. This supports treating this review as broader than the proposed simulation-driven evidence base, but it does not establish that no prior review isolated such workflows or conducted study-level extraction of autonomy, validation, and reproducibility. No page number is available in our copy. (LLM, conf=0.76)
- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
- Could not verify
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“Other reviews concentrate on natural-language processing in materials science, large language models for materials and chemistry, reticular chemistry, or generative crystal-structure modeling [5,11-14].”
What the sources say
- Supportedcites [5] Ramos, 2025, A review of large language models and autonomous agents in chemistry, Chem. Sci. source
The Introduction (‘Challenges in chemistry’; page unavailable) describes this paper as a detailed survey of LLM work in chemistry, and the Conclusions say it discusses the evolution of LLMs in chemistry and biochemistry. It therefore serves as an example of the LLM-and-chemistry reviews named in the multi-citation sentence; the other listed review topics are not attributed specifically to this source. (LLM, conf=0.92)
- Supportedcites [11] Jiang, 2025, Applications of natural language processing and large language models in materials discovery, npj Comput. Mater. source
The article is identified as a Review Article in the title information, and the Introduction says it provides an overview of NLP approaches and materials language models. That supports citing it as an example of a review focused on NLP in materials science; the other topics in the citation cluster need not be covered by this source. (LLM, conf=0.94)
- Supportedcites [12] Pei, 2025, Language models for materials discovery and sustainability: Progress, challenges, and opportunities, Prog. Mater. Sci. source
The Conclusions identify this paper as a review of natural-language processing and large language models in materials science. In this multi-citation sentence, the source serves as an example of a review in that area; it need not cover every topic listed. (LLM, conf=0.92)
- Not checked — no full text of the sourcecites [13] Zheng, 2025, Large language models for reticular chemistry, Nat. Rev. Mater. source
- Supportedcites [14] Wang, 2025, Generative AI for crystal structures: A review, npj Comput. Mater. source
The Abstract identifies this paper as a review of generative models for inorganic crystalline materials and says it organizes existing models into a taxonomy. That supports the generative crystal-structure modeling portion of the multi-citation sentence; the other review topics are not attributed specifically to this source. (LLM, conf=0.94)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The central challenge is not only to predict a property, but to manage the sequence of reasoning and action that connects a scientific objective to executable computational or experimental steps [2-10].”
What the sources say
- Supportedcites [2] Pitera, 2022, Accelerating materials discovery using artificial intelligence, high performance computing and robotics, npj Comput. Mater. source
The Introduction describes materials discovery as a sequence from research question through hypothesis formation to experimentation and testing, with bottlenecks at several stages. The simulation-workflows section describes translating a property of interest into a computational workflow, while the Outlook emphasizes orchestrating distinct capabilities across the discovery cycle. Together, these support the claim that the challenge extends beyond property prediction to connecting objectives with executable steps. (LLM, conf=0.91)
- Supportedcites [3] Tom, 2024, Self-Driving Laboratories for Chemistry and Materials Science, Chem. Rev. source
The Infrastructure > Software section (page unavailable) distinguishes property prediction from the planning and orchestration needed for discovery. It describes computational simulation workflows and the translation of sequential chemical protocols into software-executed workflows, supporting the claim’s broader point. (LLM, conf=0.86)
- Not checked — no full text of the sourcecites [4] Cheng, 2026, Artificial intelligence-driven approaches for materials design and discovery, Nat. Mater. source
- Supportedcites [5] Ramos, 2025, A review of large language models and autonomous agents in chemistry, Chem. Sci. source
The Introduction (‘Challenges in chemistry’; page not provided) identifies property prediction, molecule generation, synthesis prediction, and automation as connected challenges, and says autonomous agents connect models with other tools. ‘Agents for experiments planning’ (page not provided) describes agents that turn research tasks into experimental procedures and operate laboratory equipment. Together, these passages support the claim that discovery involves coordinating steps beyond property prediction. (LLM, conf=0.87)
- Partly supportedcites [6] Yoshikawa, 2023, Large language models for chemistry robotics, Auton. Robot. sourceExperimental — accuracy not yet measured
The Methods section describes converting natural-language experiment instructions into structured robot plans and motions, supporting the claim about connecting an objective to executable experimental steps. The paper does not frame this as a challenge beyond property prediction or address executable computational steps. (LLM, conf=0.84)
- Supportedcites [7] Bran, 2024, Augmenting large language models with chemistry tools, Nat. Mach. Intell. source
The Main section describes ChemCrow as connecting a user’s goal to an iterative sequence of reasoning, tool selection and execution across chemical tasks, including materials design. The Methods > Tools section describes how its planning and execution tools turn a synthesis objective into actionable steps. The paper thus supports the claim’s contrast between property prediction alone and managing a sequence of actions, though it does not use the phrase “central challenge.” No page numbers are available in our copy. (LLM, conf=0.87)
- Partly supportedcites [8] Boiko, 2023, Autonomous chemical research with large language models, Nature sourceExperimental — accuracy not yet measured
The Main and Integrated chemical experiment design sections describe Coscientist planning experiments, gathering information, performing calculations and writing executable code for laboratory equipment. That supports the reasoning-to-action part of the claim. The paper does not frame this as the central challenge in materials discovery or contrast it with property prediction; its examples concern chemical synthesis and reaction optimization. The supplied passages have no page numbers. (LLM, conf=0.82)
- Partly supportedcites [9] Jablonka, 2024, Leveraging large language models for predictive chemistry, Nat. Mach. Intell. sourceExperimental — accuracy not yet measured
The Classification section (page not provided) says the paper addresses questions about material properties, synthesis, and design; the Inverse design section describes generating molecules with desired properties. Neither section frames the central challenge as managing a sequence of reasoning and actions from a scientific objective to executable computational or experimental steps. (LLM, conf=0.85)
- Supportedcites [10] Jablonka, 2023, 14 examples of how LLMs can transform materials science and chemistry: A reflection on a large language model hackathon, Digit. Discov. source
The Introduction contrasts property prediction with the difficulty of using varied tools and models. Section 1.2.1 (MAPI-LLM) describes translating a scientific question into a workflow that selects databases or calculations to reach an answer; the BOLLaMa example also connects a user request to suggested experimental steps. The phrase “central challenge” is the citing authors’ framing rather than a characterization the source must state verbatim. (LLM, conf=0.87)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“A third group relied on more compact planner-executor or single-agent loops that repeatedly called tools and updated decisions from the observed outputs [24,28,29,32-34,39,40].”
What the sources say
- Supportedcites [24] Liu, 2025, Toward Automated Simulation Research Workflow through LLM Prompt Engineering Design, J. Chem. Inf. Model. source
The API Automation Program (AutoProg) section describes a single-agent loop that executes generated programs, returns outputs or errors to the LLM, and uses them to determine the next step. The Automating Agent Coordination through Sub-RP Generation section also describes a Manager AI assigning subtasks to Executor AIs and receiving their reports. Our copy gives no page numbers for these passages. (LLM, conf=0.91)
- Partly supportedcites [28] Ock, 2026, Adsorb-Agent: Autonomous identification of stable adsorption configurations via a large language model agent, Digit. Discov. sourceExperimental — accuracy not yet measured
The Initial configuration candidates section describes an iterative planner–critic loop: the Critic reviews a proposed solution and the Solution Planner revises it if needed. The Figure 1 caption also describes a framework combining LLM modules and tools. But the Energy determination section says the computational steps run without LLM involvement, so the copy does not support the more specific claim that the agent repeatedly calls tools and updates its decisions from their outputs; page numbers are unavailable. (LLM, conf=0.84)
- Not checked — no full text of the sourcecites [29] Roy, 2026, Agentic AI for autonomous discovery of radiation-tolerant CoCrFeMnNi high entropy alloys, Scr. Mater. source
- Not checked — no full text of the sourcecites [32] Baibakova, 2026, Agentic framework for programmatic crystal structure generation using a fine-tuned worker-supervisor large language model, Energy AI source
- Not checked — no full text of the sourcecites [33] Roy, 2026, Autonomous alloy composition optimization using molecular dynamics guided by a large language model, Scr. Mater. source
- Partly supportedcites [34] Wang, 2026, Catalysis AI Agent Guides Discovering the Universal Design Principle of Cu-Based Single-Atom Alloy Catalysts for CO2 Electroreduction, Angew. Chem. Int. Ed. sourceExperimental — accuracy not yet measured
The Results and discussion section describes an LLM-based Catalysis AI Agent that analyzed experimental and simulation data and helped guide the computational strategy. It does not describe a compact planner-executor or single-agent loop that repeatedly called tools and updated decisions from their outputs (searched for those terms and tool calling). No page numbers were available in our copy. (LLM, conf=0.78)
- Supportedcites [39] Wang, 2026, Accelerating materials discovery via AI-Agent integration of large language models and simulation tools, J. Mater. Inform. source
The Introduction (page not provided) describes this paper's Agent as using sequential reasoning and ReAct to invoke computational tools and dynamically refine task plans in response to intermediate results. That supports its inclusion as an example of a tool-using agent loop; the 'third group' framing belongs to the citing document. (LLM, conf=0.91)
- Not checked — no full text of the sourcecites [40] Lv, 2025, Bridging language models and computational materials science: A prompt-driven framework for material property prediction, Mater. Genome Eng. Adv. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
56 more claims · 13 with their cited sources checked
Create a free account to see all 130 claims and every source verdict.
- It may state two separate claims as one. 50
- Probably not a substantial claim. 26
- Borderline: one of the tests for a substantial claim is unclear. 7
Citations
Cites
The PRISMA 2020 statement: An updated guideline for reporting systematic reviews
Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., et al. (2021) · BMJ · doi:10.1136/bmj.n71 · Cited by 108,283
Page, M. J., Moher, D., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., et al. (2021) · BMJ · doi:10.1136/bmj.n160 · Cited by 13,262
Machine learning for molecular and materials science
Butler, K. T., Davies, D. W., Cartwright, H., Isayev, O., & Walsh, A. (2018) · Nature · doi:10.1038/s41586-018-0337-2 · Cited by 4,968
The PRISMA 2020 statement: An updated guideline for reporting systematic reviews
Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C., Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E., et al. (2021) · Journal of Clinical Epidemiology · doi:10.1016/j.jclinepi.2021.03.001 · Cited by 4,265
Autonomous chemical research with large language models
Boiko, D. A., MacKnight, R., Kline, B., & Gomes, G. (2023) · Nature · doi:10.1038/s41586-023-06792-0 · Cited by 1,050
Augmenting large language models with chemistry tools
M. Bran, A., Cox, S., Schilter, O., Baldassari, C., White, A. D., & Schwaller, P. (2024) · Nature Machine Intelligence · doi:10.1038/s42256-024-00832-8 · Cited by 794
Self-Driving Laboratories for Chemistry and Materials Science
Tom, G., Schmid, S. P., Baird, S. G., Cao, Y., Darvish, K., Hao, H., Lo, S., Pablo-García, S., Rajaonson, E. M., Skreta, M., et al. (2024) · Chemical Reviews · doi:10.1021/acs.chemrev.4c00055 · Cited by 707
Pyzer-Knapp, E. O., Pitera, J. W., Staar, P. W. J., Takeda, S., Laino, T., Sanders, D. P., Sexton, J., Smith, J. R., & Curioni, A. (2022) · npj Computational Materials · doi:10.1038/s41524-022-00765-z · Cited by 465
Leveraging large language models for predictive chemistry
Jablonka, K. M., Schwaller, P., Ortega-Guerrero, A., & Smit, B. (2024) · Nature Machine Intelligence · doi:10.1038/s42256-023-00788-1 · Cited by 378
A review of large language models and autonomous agents in chemistry
Ramos, M. C., Collison, C., & White, A. D. (2024) · Chemical Science · doi:10.48550/arxiv.2407.01603 · Cited by 277
Metadata
- Authors
- Alviz-Meza, A., Valencia-Arias, A., Rojas-Flores, S., & Diaz, F.
- Affiliations
- Escuela de Posgrado, Universidad Continental, Lima 15302, Peru
- Universidad de Los Lagos (CL)
- Universidad César Vallejo (PE)
- Universidad Autónoma del Perú (PE)
- Venue
- Data
- Publisher
- MDPI AG
- Published
- 2026-07-21
- Volume
- 11
- Issue
- 7
- Pages
- 180
- ISSN
- 2306-5729 (ISSN-L)
- OpenAlex
- W7169861758
- Semantic Scholar
- 969b895a2fd689bcc40271df058ca5095005ee23
- Cited by
- 0
- References
- 41
- FWCI
- 0.00
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Data (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- Topics
- Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
- Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence Applications
- Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
- Keywords
- AI agents, LLM agents, materials discovery, agentic reasoning, workflow orchestration, multi-agent architecture, materials informatics, self-driving laboratories, molecular dynamics simulation, alloy design, catalysis, adsorption, reproducibility, research infrastructure
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-08, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-08 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.