CoChat

14 examples of how LLMs can transform materials science and chemistry: a reflection on a large language model hackathon

Open access

The DOI matches its Crossref record; 109 of the 109 references it cites were found in a registry or an index.

Abstract

We report the findings of a hackathon focused on exploring the diverse applications of large language models in molecular and materials science.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1039/d3dd00113j

Cites

109 of 109 references verified0 citation errors · 0 not checkable

Statistics

Figures

Image 1
CC-BY · From the PDF
Image 2
CC-BY · From the PDF
Fig. 1 Using LLMs to predict the compressive strength of concretes. An illustration of the conventional approach for solving this task, i.e. , training classical prediction models using ten training data points as tabular data (left). Using the LIFT framework LLMs can also use tabular data and leverage context information provided in natural language (right). The context can be ' fuzzy ' design rules often known in chemistry and materials science but hard to incorporate in conventional ML models. Augmented with this context and ten training examples, ICL with LLM leads to a performance that outperforms baselines such as RFs or GPR.
Fig. 1 Using LLMs to predict the compressive strength of concretes. An illustration of the conventional approach for solving this task, i.e. , training classical prediction models using ten training data points as tabular data (left). Using the LIFT framework LLMs can also use tabular data and leverage context information provided in natural language (right). The context can be ' fuzzy ' design rules often known in chemistry and materials science but hard to incorporate in conventional ML models. Augmented with this context and ten training examples, ICL with LLM leads to a performance that outperforms baselines such as RFs or GPR.CC-BY · From the PDF
Fig. 2 GA using an LLM. This fi gure illustrates how di ff erent aspects of a GA can be performed by an LLM. GPT-3.5 was used to fragment, reproduce, and optimize molecules represented by SMILES strings. The fi rst column illustrated how an LLM can fragment a molecule represented by a SMILES string (input molecule on top, output LLM fragments below). The middle column showcases how an LLM can reproduce/mix two molecules as is done in a GA (input molecule on top, output LLM below). The right column illustrates an application in which an LLM is used to optimize molecules given their SMILES and an associated score. The LLM suggested potential modi fi cations to optimize molecules. The plot shows best (blue) and mean (orange) Tanimoto similarity to vitamin C per LLM produced generations.
Fig. 2 GA using an LLM. This fi gure illustrates how di ff erent aspects of a GA can be performed by an LLM. GPT-3.5 was used to fragment, reproduce, and optimize molecules represented by SMILES strings. The fi rst column illustrated how an LLM can fragment a molecule represented by a SMILES string (input molecule on top, output LLM fragments below). The middle column showcases how an LLM can reproduce/mix two molecules as is done in a GA (input molecule on top, output LLM below). The right column illustrates an application in which an LLM is used to optimize molecules given their SMILES and an associated score. The LLM suggested potential modi fi cations to optimize molecules. The plot shows best (blue) and mean (orange) Tanimoto similarity to vitamin C per LLM produced generations.CC-BY · From the PDF
Fig. 3 Schematic overview of the MAPI-LLM work fl ow. It uses LLMs to process the user's input and decide which available tools ( e.g. , Materials Project API, the Reaction-Network package, and Google Search) to use following an iterative chain-of-thought procedure. In this way, it can answer questions such as ' Is the material AnByCz stable? ' .
Fig. 3 Schematic overview of the MAPI-LLM work fl ow. It uses LLMs to process the user's input and decide which available tools ( e.g. , Materials Project API, the Reaction-Network package, and Google Search) to use following an iterative chain-of-thought procedure. In this way, it can answer questions such as ' Is the material AnByCz stable? ' .CC-BY · From the PDF
Fig. 4 The sMolTalk interface. Based on few-shot prompting LLMs can create code for visualization tools such as that can create custom visualization based on a natural-language description of the desired output. The top left box is the input fi eld where users can enter commands in natural language. The top right box prints the code the LLM generates. This code generates the visualization shown in the lower box. In this example, the user entered a sequence of four commands: the LLM (1) generates code for retrieving the structure, (2) colors the carbons blue, (3) displays the hydrogens as red spheres, and (4) reduces the size of the spheres.
Fig. 4 The sMolTalk interface. Based on few-shot prompting LLMs can create code for visualization tools such as that can create custom visualization based on a natural-language description of the desired output. The top left box is the input fi eld where users can enter commands in natural language. The top right box prints the code the LLM generates. This code generates the visualization shown in the lower box. In this example, the user entered a sequence of four commands: the LLM (1) generates code for retrieving the structure, (2) colors the carbons blue, (3) displays the hydrogens as red spheres, and (4) reduces the size of the spheres.CC-BY · From the PDF
Fig. 5 Using an LLM as an interface to an ELN/data management system/data management system. LLM-based assistants can provide powerful interfaces to digital experimental data. The fi gure shows a screenshot of a conversation with in the data management system ( https://github.com/the-grey-group/datalab ). Here, is provided with data from the JSON API of of an experimental battery cell. The user then prompts (green box) the system to build a fl owchart of the provenance of the sample. The assistant responds with markdown code, which the interface automatically recognizes and translates into a visualization.
Fig. 5 Using an LLM as an interface to an ELN/data management system/data management system. LLM-based assistants can provide powerful interfaces to digital experimental data. The fi gure shows a screenshot of a conversation with in the data management system ( https://github.com/the-grey-group/datalab ). Here, is provided with data from the JSON API of of an experimental battery cell. The user then prompts (green box) the system to build a fl owchart of the provenance of the sample. The assistant responds with markdown code, which the interface automatically recognizes and translates into a visualization.CC-BY · From the PDF
Fig. 6 Schematic overview of BoLLama. An LLM can act as an interface to a BO algorithm. An experimental chemist can bootstrap an optimization and then, via a chat interface, update the state of the simulation to which the bot responds with the recommended next steps.
Fig. 6 Schematic overview of BoLLama. An LLM can act as an interface to a BO algorithm. An experimental chemist can bootstrap an optimization and then, via a chat interface, update the state of the simulation to which the bot responds with the recommended next steps.CC-BY · From the PDF
Fig. 7 The InsightGraph interface. A suitably prompted LLM can create knowledge graph representations of scienti fi c text that can be visualized using tools such as neo4j's visualization tools. 83
Fig. 7 The InsightGraph interface. A suitably prompted LLM can create knowledge graph representations of scienti fi c text that can be visualized using tools such as neo4j's visualization tools. 83CC-BY · From the PDF
Fig. 9 TableToJson. Results of the structured JSON generation of tables contained in scienti fi c articles. Two approaches are compared: (i) the use of an OpenAI model prompted with the desired JSON schema, and (ii) the use of an OpenAI model together with In both cases, JSON objects were always obtained. The output of the OpenAI model did not always follow the provided schema, although this might be solved by modifying the schema. The accuracy of the results from the approach used with OpenAI models could be increased (as shown by the blue arrows) by solving errors in the generation of power numbers and special characters with a more detailed prompt. The results can be visualized in this demo app: https://vgvinter-tabletojson-app-kt5aiv.streamlit.app/ .
Fig. 9 TableToJson. Results of the structured JSON generation of tables contained in scienti fi c articles. Two approaches are compared: (i) the use of an OpenAI model prompted with the desired JSON schema, and (ii) the use of an OpenAI model together with In both cases, JSON objects were always obtained. The output of the OpenAI model did not always follow the provided schema, although this might be solved by modifying the schema. The accuracy of the results from the approach used with OpenAI models could be increased (as shown by the blue arrows) by solving errors in the generation of power numbers and special characters with a more detailed prompt. The results can be visualized in this demo app: https://vgvinter-tabletojson-app-kt5aiv.streamlit.app/ .CC-BY · From the PDF
Image 11
CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 109 references

Research works printed but not linked: 11

  1. Vaswani, 2017, Adv. Neural Inf. Process. Syst., 30, 6000
  2. Brown, 2020, Adv. Neural Inf. Process. Syst., 33, 1877
  3. C. N.Edwards , T.Lai , K.Ros , G.Honke and H.Ji , Translation between Molecules and Natural Language , Conference On Empirical Methods In Natural Language Processing , 2022
  4. Krenn, 2020, Mach. Learn.: Sci. Technol., 1, 045024
  5. Ramakrishnan, 2015, J. Chem. Theory Comput., 11, 2087, 10.1021/acs.jctc.5b00099 · DOI · Paper page
  6. A.Radford , J.Wu , R.Child , D.Luan , D.Amodei and I.Sutskever , Language Models are Unsupervised Multitask Learners , 2019 , https://d4mucfpksywv.cloudfront.net/better-language-models/language_models_are_unsupervised_multitask_learners.pdf
  7. T.Mikolov , K.Chen , G.Corrado and J.Dean , Efficient Estimation of Word Representations in Vector Space , International Conference On Learning Representations , 2013
  8. Wei, 2022, Neural. Inf. Process. Syst., 24824
  9. Neo4j , Neo4j – The World's Leading Graph Database , 2012 , http://neo4j.org/
  10. Raffel, 2020, J. Mach. Learn. Res., 21, 5485
  11. Editorials, 2023, Nature, 613, 10

Cited by

Items 101–200 of 251

Metadata

Authors
Jablonka, K. M., Ai, Q., Al-Feghali, A., Badhwar, S., Bocarsly, J. D., Bran, A. M., Bringuier, S., Brinson, L. C., Choudhary, K., Circi, D., Cox, S., de Jong, W. A., Evans, M. L., Gastellu, N., Genzling, J., Gil, M. V., Gupta, A. K., Hong, Z., Imran, A., Kruschwitz, S., Labarre, A., Lála, J., Liu, T., Ma, S., Majumdar, S., Merz, G. W., Moitessier, N., Moubarak, E., Mouriño, B., Pelkie, B., Pieler, M., Ramos, M. C., Ranković, B., Rodriques, S. G., Sanders, J. N., Schwaller, P., Schwarting, M., Shi, J., Smit, B., Smith, B. E., Van Herck, J., Völker, C., Ward, L., Warren, S., Weiser, B., Zhang, S., Zhang, X., Zia, G. A., Scourtas, A., Schmidt, K. J., Foster, I., White, A. D., & Blaiszik, B.
Affiliations
  • École Polytechnique Fédérale de Lausanne (CH)
  • Massachusetts Institute of Technology (US)
  • Department of Chemistry, McGill University, Montreal, Quebec, Canada
  • Reincarnate Inc., USA
  • University of Cambridge (GB)
  • NCCR Catalysis (CH)
  • Independent Researcher, San Diego, CA, USA
  • Duke University (US)
  • National Institute of Standards and Technology (US)
  • Mechanical Engineering and Materials Science, Duke University, USA
  • University of Rochester (US)
  • Applied Mathematics and Computational Research Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA
  • UCLouvain (BE)
  • McGill University (CA)
  • Consejo Superior de Investigaciones Científicas (ES)
  • Lawrence Berkeley National Laboratory (US)
  • University of Chicago (US)
  • University of California, Berkeley (US)
  • Federal Institute For Materials Research and Testing (DE)
  • The Francis Crick Institute (GB)
  • University of Wisconsin–Madison (US)
  • University of Washington (US)
  • OpenBioML.org, UK
  • University of California, Los Angeles (US)
  • Laboratory of Molecular Simulation (LSMO), Institut des Sciences et Ingénierie Chimiques, Ecole Polytechnique Fédérale de Lausanne (EPFL), Sion, Valais, Switzerland
  • Argonne National Laboratory (US)
Corresponding author
  • Jablonka, Kevin Maik
  • Blaiszik, Ben
Venue
Digital Discovery
Published
2023
Volume
2
Issue
5
Pages
1233-1250
ISSN
2635-098X (ISSN-L)
OpenAlex
W4385671288
Cited by
218
References
89
Citations by year
2026: 52 · 2025: 68 · 2024: 93 · 2023: 4
FWCI
3.35
Type
journal article
Language
en
Open access
Locations
  • Digital Discovery (publishedVersion, cc-by)
  • DIGITAL.CSIC (Spanish National Research Council (CSIC)) (submittedVersion, cc-by)
  • eScholarship (California Digital Library) (submittedVersion, cc-by-nc)
  • Infoscience (Ecole Polytechnique Fédérale de Lausanne) (submittedVersion, cc-by-nc-nd)
  • OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion)
Funders
  • National Institute of General Medical Sciences (R35GM137966)
  • High Energy Physics (DE-FOA-0002705)
  • NCCR Catalysis (180544)
  • Agencia Estatal de Investigación (TED2021-131693B-I00)
  • National Institutes of Health (U18TR004149)
  • Horizon 2020 Framework Programme (957189)
  • Office of Science (DE-AC02-05CH11231)
  • Center for Hierarchical Materials Design (70NANB19H005)
  • Consejo Superior de Investigaciones Científicas (LINKA20412)
  • National Science Foundation (#2226419, #2209892, #1931306, 1764415, 1917340, DGE-2022040)
  • U.S. Department of Commerce
  • Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (51NF40-182892)
  • Grantham Foundation for the Protection of the Environment
  • Engineering and Physical Sciences Research Council
  • U.S. Department of Energy
  • Francis Crick Institute
  • Cancer Research UK
  • Medical Research Council
  • Wellcome Trust
  • H2020 Marie Skłodowska-Curie Actions (945363, 847587)
  • European Commission (RYC-2017-21937)
  • Esperantic Studies Foundation
  • Fédération Wallonie-Bruxelles
  • National Institute of Standards and Technology
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
large language models, materials science, chemistry, molecular science, hackathons

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.

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.