PolyFusionAgent: A Multimodal Foundation Model and Autonomous AI Assistant for Polymer Property Prediction and Inverse Design

Manpreet Kaur; Qian Liu; Xingying Zhang

arxiv: 2605.26543 · v1 · pith:WTRGUS7Dnew · submitted 2026-05-26 · 💻 cs.AI · cs.LG

PolyFusionAgent: A Multimodal Foundation Model and Autonomous AI Assistant for Polymer Property Prediction and Inverse Design

Manpreet Kaur , Xingying Zhang , Qian Liu This is my paper

Pith reviewed 2026-06-29 18:34 UTC · model grok-4.3

classification 💻 cs.AI cs.LG

keywords polymer discoverymultimodal foundation modellatent spaceproperty predictioninverse designPolyFusionPolyAgentthermophysical properties

0 comments

The pith

PolyFusion aligns sequence, topology, 3D geometry, and fingerprints of millions of polymers into a shared latent space to improve thermophysical property prediction and enable generation of novel valid polymers.

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

Polymer discovery faces an enormous design space and fragmented data on structure and properties. The paper presents PolyFusion, a foundation model that fuses four complementary polymer representations—sequence, topology, 3D geometry, and fingerprints—across millions of examples. This fusion creates a single latent space that transfers across different polymer chemistries and data sizes. The aligned space raises accuracy on thermophysical property prediction and supports generation of new polymers that meet target properties while remaining chemically valid. An accompanying agent called PolyAgent then ties the outputs to literature evidence so the whole process stays grounded in existing knowledge.

Core claim

PolyFusion aligns complementary polymer views including sequence, topology, 3D geometry, and fingerprints across millions of polymers to learn a shared latent space transferable across chemistries and data regimes, improving thermophysical property prediction and enabling property-conditioned generation of chemically valid, structurally novel polymers beyond the reference design space.

What carries the argument

Multimodal alignment of sequence, topology, 3D geometry, and fingerprints into one shared latent space inside PolyFusion.

If this is right

Thermophysical property predictions become more accurate across varied polymer families and data regimes.
Property-conditioned generation produces chemically valid polymers that lie outside the original training structures.
The shared latent space transfers to new chemistries without retraining from scratch.
PolyAgent links each prediction and design step to explicit literature precedent in a single workflow.

Where Pith is reading between the lines

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

The same alignment technique could shorten the cycle from computational proposal to laboratory testing in energy-storage and biomedical polymer work.
A latent space built this way might later serve as a starting point for other molecular classes such as small organics or inorganic materials.
Interactive retrieval of prior experimental results could reduce the number of designs that reach the bench without supporting evidence.

Load-bearing premise

Aligning the four polymer representations will create a latent space that generalizes to new chemistries and produces chemically valid novel polymers.

What would settle it

Synthesizing and measuring a batch of PolyFusion-generated polymers that fail to match the predicted properties or prove chemically invalid would disprove the central claim.

read the original abstract

Polymer discovery is central to fields ranging from energy storage to biomedicine, but it is hindered by an astronomically large chemical design space and fragmented representations of structure, properties, and prior knowledge. This fragmentation leaves many AI models disconnected from physical and experimental reality, restricting their ability to support directly actionable design decisions. Here we introduce PolyFusionAgent, an interactive framework coupling a multimodal polymer foundation model (PolyFusion) with a tool-augmented, literature-grounded design agent (PolyAgent). PolyFusion aligns complementary polymer views including sequence, topology, 3D geometry, and fingerprints across millions of polymers to learn a shared latent space transferable across chemistries and data regimes, improving thermophysical property prediction and enabling property-conditioned generation of chemically valid, structurally novel polymers beyond the reference design space. PolyAgent closes the design loop by linking prediction and inverse design with evidence retrieval from the polymer literature, proposing, evaluating, and contextualizing hypotheses with explicit precedent in one workflow. Together, PolyFusionAgent enables interactive, evidence-linked polymer discovery combining large-scale representation learning, multimodal chemical knowledge, and verifiable scientific reasoning.

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

PolyFusionAgent proposes aligning four polymer representations into a shared latent space and pairing it with a literature-retrieving agent, but the abstract supplies no metrics or methods to check whether the claimed gains in prediction or generation actually occur.

read the letter

The main point is that this paper describes PolyFusion, a multimodal model that fuses sequence, topology, 3D geometry, and fingerprints across millions of polymers, plus PolyAgent, which pulls literature evidence into the design loop. The stated goal is better thermophysical predictions and property-conditioned generation of chemically valid new polymers.

The framework idea is straightforward and addresses a real issue: polymer data comes in disconnected forms, and many models lack ties to experimental records. Linking the model outputs to explicit literature precedent through the agent is a practical step that could make inverse design more usable in energy and biomedicine contexts.

The limitation is that nothing in the abstract shows results, baselines, ablation studies, or validation of chemical validity. The claim that the latent space transfers across chemistries and data regimes and produces structures beyond the training distribution is asserted without evidence or error analysis. The weakest assumption listed in the report—that multimodal alignment will deliver generalization and valid novel polymers—remains untested here. Without the full methods or data, it is impossible to tell whether the approach adds anything beyond standard multimodal representation learning.

This is aimed at researchers working on AI tools for polymer discovery who want to combine foundation models with agentic retrieval. A reader focused on multimodal chemistry or closing the loop between prediction and literature might extract useful framing, but the lack of quantitative support limits immediate applicability.

The paper deserves peer review because the integration of multiple polymer views with evidence-linked agents is a direction worth examining in detail, even if the current version is high-level. Send it for refereeing to assess the implementation and experiments.

Referee Report

2 major / 0 minor

Summary. The manuscript introduces PolyFusionAgent, an interactive framework that couples a multimodal polymer foundation model (PolyFusion) with a tool-augmented, literature-grounded design agent (PolyAgent). PolyFusion is described as aligning sequence, topology, 3D geometry, and fingerprints across millions of polymers to learn a shared latent space that is transferable across chemistries and data regimes, thereby improving thermophysical property prediction and enabling property-conditioned generation of chemically valid, structurally novel polymers. PolyAgent is presented as closing the design loop by linking prediction and inverse design with evidence retrieval from the polymer literature to propose, evaluate, and contextualize hypotheses.

Significance. If the multimodal alignment produces a genuinely transferable latent space that generalizes beyond the training distribution and the agent workflow yields verifiable, literature-grounded designs, the work could meaningfully advance AI-assisted polymer discovery by unifying fragmented structural representations and supporting closed-loop, evidence-linked design in application areas such as energy storage and biomedicine.

major comments (2)

[Abstract] Abstract: the central claims of improved thermophysical property prediction and generation of chemically valid novel polymers beyond the reference design space are stated without any reported metrics, baselines, validation protocols, or error analysis, rendering it impossible to assess whether the multimodal alignment delivers the asserted performance gains or generalization.
[Abstract] Abstract: the assumption that alignment across sequence, topology, 3D geometry, and fingerprints will produce a latent space transferable across chemistries and data regimes is presented as a core contribution, yet no implementation details, loss functions, alignment objectives, or cross-chemistry transfer experiments are supplied to support or refute this claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for highlighting issues in the abstract that limit immediate assessment of our claims. We agree the abstract should be more self-contained and will revise it to incorporate concise quantitative summaries and technical pointers drawn from the full manuscript, without altering the underlying results.

read point-by-point responses

Referee: [Abstract] Abstract: the central claims of improved thermophysical property prediction and generation of chemically valid novel polymers beyond the reference design space are stated without any reported metrics, baselines, validation protocols, or error analysis, rendering it impossible to assess whether the multimodal alignment delivers the asserted performance gains or generalization.

Authors: We agree the abstract would benefit from explicit metrics to support the claims. The full manuscript contains comparative tables, baseline results, MAE/RMSE values, validity percentages, and out-of-distribution transfer metrics in Sections 3 and 4, along with the validation protocols used. In revision we will add a single sentence to the abstract that summarizes the key quantitative gains (e.g., relative improvement ranges and validity rates) while directing readers to the detailed tables and protocols in the body. revision: yes
Referee: [Abstract] Abstract: the assumption that alignment across sequence, topology, 3D geometry, and fingerprints will produce a latent space transferable across chemistries and data regimes is presented as a core contribution, yet no implementation details, loss functions, alignment objectives, or cross-chemistry transfer experiments are supplied to support or refute this claim.

Authors: The abstract states the high-level claim, but the manuscript supplies the requested details: the multimodal contrastive alignment objectives and loss functions are defined in Section 2.2, the training regime across modalities is described in Section 2.3, and cross-chemistry transfer experiments (including held-out polymer families) appear in Section 3.2 with associated figures. We will revise the abstract to include a brief clause referencing the contrastive alignment approach and the existence of transfer experiments, while retaining the full technical exposition in the methods and results. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified from available text

full rationale

The abstract and provided description frame PolyFusion as a multimodal alignment procedure across sequence, topology, 3D geometry, and fingerprints to produce a shared latent space. This is a standard representation-learning statement with no equations, self-citations, fitted parameters renamed as predictions, or self-definitional steps visible. No load-bearing claim reduces to its own inputs by construction. The central claim remains independent of the listed circularity patterns. Honest non-finding applies when the text supplies no quotable reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only abstract available; no information on free parameters, axioms, or invented entities is extractable.

pith-pipeline@v0.9.1-grok · 5730 in / 1053 out tokens · 25091 ms · 2026-06-29T18:34:29.163163+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

27 extracted references · 19 canonical work pages

[1]

Basic concepts and tools of artificial intelligence in polymer science.Polymer Chemistry16, 2457–2470 (2025)

Ferji, K. Basic concepts and tools of artificial intelligence in polymer science.Polymer Chemistry16, 2457–2470 (2025). URL https://doi.org/10.1039/D5PY00148J. 12

work page doi:10.1039/d5py00148j 2025
[2]

& Smith, J

Takeda, S., Kishimoto, A., Hamada, L., Nakano, D. & Smith, J. R.Foundation model for material science, Vol. 37, 15376–15383 (2023). URL https://doi.org/10.1609/aaai.v37i13. 26793

work page doi:10.1609/aaai.v37i13 2023
[3]

URL https://doi.org/10.1021/ acscentsci.9b00476

Lin, T.-S.et al.Bigsmiles: A structurally-based line notation for describing macro- molecules.ACS Central Science5, 1523–1531 (2019). URL https://doi.org/10.1021/ acscentsci.9b00476

2019
[4]

Self- referencing embedded strings (SELFIES): A 100% robust molecular string representation.Machine Learning: Science and Technology, 1(4):045024, 2020

Krenn, M., H¨ ase, F., Nigam, A., Friederich, P. & Aspuru-Guzik, A. Self-referencing embed- ded strings (SELFIES): A 100% robust molecular string representation.Machine Learning: Science and Technology1, 045024 (2020). URL https://doi.org/10.1088/2632-2153/aba947

work page doi:10.1088/2632-2153/aba947 2020
[5]

Uni-Mol: A universal 3d molecular representation learning framework (2023)

Zhou, G.et al. Uni-Mol: A universal 3d molecular representation learning framework (2023). URL https://openreview.net/forum?id=6K2RM6wVqKu

2023
[6]

& Barati Farimani, A

Wang, Y., Wang, J., Cao, Z. & Barati Farimani, A. Molecular contrastive learning of representations via graph neural networks.Nature Machine Intelligence4, 279–287 (2022). URL https://doi.org/10.1038/s42256-022-00447-x

work page doi:10.1038/s42256-022-00447-x 2022
[7]

Learning transferable visual models from natural language supervision, Vol

Radford, A.et al. Learning transferable visual models from natural language supervision, Vol. 139 ofProceedings of Machine Learning Research, 8748–8763 (PMLR, 2021). URL https://proceedings.mlr.press/v139/radford21a.html

2021
[8]

The Journal of Physical Chem- istry C120(40), 23111–23120 (2016)

Kim, C., Chandrasekaran, A., Huan, T. D., Das, D. & Ramprasad, R. Polymer genome: A data-powered polymer informatics platform for property predictions.The Journal of Physical Chemistry C122, 17575–17585 (2018). URL https://doi.org/10.1021/acs.jpcc. 8b02913

work page doi:10.1021/acs.jpcc 2018
[9]

A general-purpose machine learning framework for predicting properties of inorganic materials

Ward, L., Agrawal, A., Choudhary, A. & Wolverton, C. A general-purpose machine learning framework for predicting properties of inorganic materials.npj Computational Materials2, 16028 (2016). URL https://doi.org/10.1038/npjcompumats.2016.28

work page doi:10.1038/npjcompumats.2016.28 2016
[10]

& Jaakkola, T.Junction tree variational autoencoder for molecular graph generation, Vol

Jin, W., Barzilay, R. & Jaakkola, T.Junction tree variational autoencoder for molecular graph generation, Vol. 80 ofProceedings of Machine Learning Research, 2323–2332 (PMLR, 2018). URL https://proceedings.mlr.press/v80/jin18a.html

2018
[11]

& Kim, C

Ramprasad, R., Batra, R., Pilania, G., Mannodi-Kanakkithodi, A. & Kim, C. Machine learning in materials informatics: recent applications and prospects.npj Computational Materials3, 54 (2017). URL https://doi.org/10.1038/s41524-017-0056-5

work page doi:10.1038/s41524-017-0056-5 2017
[12]

& Ramprasad, R

Pilania, G., Wang, C., Jiang, X., Rajasekaran, S. & Ramprasad, R. Accelerating materials property predictions using machine learning.Scientific Reports3, 2810 (2013). URL https: //doi.org/10.1038/srep02810

work page doi:10.1038/srep02810 2013
[13]

& Ramprasad, R

Kuenneth, C. & Ramprasad, R. polybert: a chemical language model to enable fully machine-driven ultrafast polymer informatics.Nature Communications14, 4099 (2023). URL https://doi.org/10.1038/s41467-023-39868-6

work page doi:10.1038/s41467-023-39868-6 2023
[14]

& Ramprasad, R

Kuenneth, C. & Ramprasad, R. polyOne data set – 100 million hypothetical polymers including 29 properties (2022). URL https://doi.org/10.5281/zenodo.7766806

work page doi:10.5281/zenodo.7766806 2022
[15]

& Barati Farimani, A

Xu, C., Wang, Y. & Barati Farimani, A. Transpolymer: a transformer-based language model for polymer property predictions.npj Computational Materials9, 64 (2023). URL https://doi.org/10.1038/s41524-023-01016-5

work page doi:10.1038/s41524-023-01016-5 2023
[16]

URL https://doi.org/10.1039/ D3SC05079C

Qiu, H.et al.Polync: a natural and chemical language model for the prediction of unified polymer properties.Chemical Science15, 534–544 (2024). URL https://doi.org/10.1039/ D3SC05079C

2024
[17]

Mmpolymer: A multimodal multitask pretraining framework for poly- mer property prediction, 2336–2346 (ACM, 2024)

Wang, F.et al. Mmpolymer: A multimodal multitask pretraining framework for poly- mer property prediction, 2336–2346 (ACM, 2024). URL https://doi.org/10.1145/3627673. 3679684. 13

work page doi:10.1145/3627673 2024
[18]

URL https://doi.org/10.1038/ s41524-025-01652-z

Huang, Q.et al.Unified multimodal multidomain polymer representation for property prediction.npj Computational Materials11, 153 (2025). URL https://doi.org/10.1038/ s41524-025-01652-z

2025
[19]

& Sun, Z.-Y

Qiu, H. & Sun, Z.-Y. On-demand reverse design of polymers with polytao.npj Computa- tional Materials10, 273 (2024). URL https://doi.org/10.1038/s41524-024-01466-5

work page doi:10.1038/s41524-024-01466-5 2024
[20]

S., Xiong, W

Savit, A., Sahu, H., Shukla, S. S., Xiong, W. & Ramprasad, R.polybart: a chemical lin- guist for polymer property prediction and generative design, 12104–12119 (Association for Computational Linguistics, 2025). URL https://doi.org/10.18653/v1/2025.findings-emnlp. 647

work page doi:10.18653/v1/2025.findings-emnlp 2025
[21]

Sahu, H., Xiong, W., Savit, A., Shukla, S. S. & Ramprasad, R. Polyt5: an encoder- decoder foundation chemical language model for generative polymer design.npj Artificial Intelligence2, 30 (2026). URL https://doi.org/10.1038/s44387-026-00087-1

work page doi:10.1038/s44387-026-00087-1 2026
[22]

URL https: //doi.org/10.1021/acs.chemmater.1c02061

Gurnani, R.et al.polyg2g: A novel machine learning algorithm applied to the generative design of polymer dielectrics.Chemistry of Materials33, 7008–7016 (2021). URL https: //doi.org/10.1021/acs.chemmater.1c02061

work page doi:10.1021/acs.chemmater.1c02061 2021
[23]

& Wang, X

Li, W., Li, Y., Lei, Q., Wang, Z. & Wang, X. PolyRL: reinforcement learning-guided polymer generation for multi-objective polymer discovery.Digital Discovery5, 266–276 (2026). URL https://doi.org/10.1039/D5DD00272A

work page doi:10.1039/d5dd00272a 2026
[24]

Yue, T., Tao, L., Varshney, V. & Li, Y. Benchmarking study of deep generative models for inverse polymer design.Digital Discovery4, 910–926 (2025). URL https://doi.org/10. 1039/D4DD00395K

2025
[25]

& Luo, T

Ma, R. & Luo, T. PI1M: A benchmark database for polymer informatics.Journal of Chemical Information and Modeling60, 4684–4690 (2020). URL https://doi.org/10.1021/ acs.jcim.0c00726

2020
[26]

& Yamazaki, M.PoLyInfo: Polymer database for polymeric materials design, 22–29 (IEEE, 2011)

Otsuka, S., Kuwajima, I., Hosoya, J., Xu, Y. & Yamazaki, M.PoLyInfo: Polymer database for polymeric materials design, 22–29 (IEEE, 2011). URL https://doi.org/10.1109/EIDWT. 2011.13

work page doi:10.1109/eidwt 2011
[27]

URL https: //doi.org/10.1038/s43018-025-00991-6

Ferber, D.et al.Development and validation of an autonomous artificial intelligence agent for clinical decision-making in oncology.Nature Cancer6, 1337–1349 (2025). URL https: //doi.org/10.1038/s43018-025-00991-6. 14

work page doi:10.1038/s43018-025-00991-6 2025

[1] [1]

Basic concepts and tools of artificial intelligence in polymer science.Polymer Chemistry16, 2457–2470 (2025)

Ferji, K. Basic concepts and tools of artificial intelligence in polymer science.Polymer Chemistry16, 2457–2470 (2025). URL https://doi.org/10.1039/D5PY00148J. 12

work page doi:10.1039/d5py00148j 2025

[2] [2]

& Smith, J

Takeda, S., Kishimoto, A., Hamada, L., Nakano, D. & Smith, J. R.Foundation model for material science, Vol. 37, 15376–15383 (2023). URL https://doi.org/10.1609/aaai.v37i13. 26793

work page doi:10.1609/aaai.v37i13 2023

[3] [3]

URL https://doi.org/10.1021/ acscentsci.9b00476

Lin, T.-S.et al.Bigsmiles: A structurally-based line notation for describing macro- molecules.ACS Central Science5, 1523–1531 (2019). URL https://doi.org/10.1021/ acscentsci.9b00476

2019

[4] [4]

Self- referencing embedded strings (SELFIES): A 100% robust molecular string representation.Machine Learning: Science and Technology, 1(4):045024, 2020

Krenn, M., H¨ ase, F., Nigam, A., Friederich, P. & Aspuru-Guzik, A. Self-referencing embed- ded strings (SELFIES): A 100% robust molecular string representation.Machine Learning: Science and Technology1, 045024 (2020). URL https://doi.org/10.1088/2632-2153/aba947

work page doi:10.1088/2632-2153/aba947 2020

[5] [5]

Uni-Mol: A universal 3d molecular representation learning framework (2023)

Zhou, G.et al. Uni-Mol: A universal 3d molecular representation learning framework (2023). URL https://openreview.net/forum?id=6K2RM6wVqKu

2023

[6] [6]

& Barati Farimani, A

Wang, Y., Wang, J., Cao, Z. & Barati Farimani, A. Molecular contrastive learning of representations via graph neural networks.Nature Machine Intelligence4, 279–287 (2022). URL https://doi.org/10.1038/s42256-022-00447-x

work page doi:10.1038/s42256-022-00447-x 2022

[7] [7]

Learning transferable visual models from natural language supervision, Vol

Radford, A.et al. Learning transferable visual models from natural language supervision, Vol. 139 ofProceedings of Machine Learning Research, 8748–8763 (PMLR, 2021). URL https://proceedings.mlr.press/v139/radford21a.html

2021

[8] [8]

The Journal of Physical Chem- istry C120(40), 23111–23120 (2016)

Kim, C., Chandrasekaran, A., Huan, T. D., Das, D. & Ramprasad, R. Polymer genome: A data-powered polymer informatics platform for property predictions.The Journal of Physical Chemistry C122, 17575–17585 (2018). URL https://doi.org/10.1021/acs.jpcc. 8b02913

work page doi:10.1021/acs.jpcc 2018

[9] [9]

A general-purpose machine learning framework for predicting properties of inorganic materials

Ward, L., Agrawal, A., Choudhary, A. & Wolverton, C. A general-purpose machine learning framework for predicting properties of inorganic materials.npj Computational Materials2, 16028 (2016). URL https://doi.org/10.1038/npjcompumats.2016.28

work page doi:10.1038/npjcompumats.2016.28 2016

[10] [10]

& Jaakkola, T.Junction tree variational autoencoder for molecular graph generation, Vol

Jin, W., Barzilay, R. & Jaakkola, T.Junction tree variational autoencoder for molecular graph generation, Vol. 80 ofProceedings of Machine Learning Research, 2323–2332 (PMLR, 2018). URL https://proceedings.mlr.press/v80/jin18a.html

2018

[11] [11]

& Kim, C

Ramprasad, R., Batra, R., Pilania, G., Mannodi-Kanakkithodi, A. & Kim, C. Machine learning in materials informatics: recent applications and prospects.npj Computational Materials3, 54 (2017). URL https://doi.org/10.1038/s41524-017-0056-5

work page doi:10.1038/s41524-017-0056-5 2017

[12] [12]

& Ramprasad, R

Pilania, G., Wang, C., Jiang, X., Rajasekaran, S. & Ramprasad, R. Accelerating materials property predictions using machine learning.Scientific Reports3, 2810 (2013). URL https: //doi.org/10.1038/srep02810

work page doi:10.1038/srep02810 2013

[13] [13]

& Ramprasad, R

Kuenneth, C. & Ramprasad, R. polybert: a chemical language model to enable fully machine-driven ultrafast polymer informatics.Nature Communications14, 4099 (2023). URL https://doi.org/10.1038/s41467-023-39868-6

work page doi:10.1038/s41467-023-39868-6 2023

[14] [14]

& Ramprasad, R

Kuenneth, C. & Ramprasad, R. polyOne data set – 100 million hypothetical polymers including 29 properties (2022). URL https://doi.org/10.5281/zenodo.7766806

work page doi:10.5281/zenodo.7766806 2022

[15] [15]

& Barati Farimani, A

Xu, C., Wang, Y. & Barati Farimani, A. Transpolymer: a transformer-based language model for polymer property predictions.npj Computational Materials9, 64 (2023). URL https://doi.org/10.1038/s41524-023-01016-5

work page doi:10.1038/s41524-023-01016-5 2023

[16] [16]

URL https://doi.org/10.1039/ D3SC05079C

Qiu, H.et al.Polync: a natural and chemical language model for the prediction of unified polymer properties.Chemical Science15, 534–544 (2024). URL https://doi.org/10.1039/ D3SC05079C

2024

[17] [17]

Mmpolymer: A multimodal multitask pretraining framework for poly- mer property prediction, 2336–2346 (ACM, 2024)

Wang, F.et al. Mmpolymer: A multimodal multitask pretraining framework for poly- mer property prediction, 2336–2346 (ACM, 2024). URL https://doi.org/10.1145/3627673. 3679684. 13

work page doi:10.1145/3627673 2024

[18] [18]

URL https://doi.org/10.1038/ s41524-025-01652-z

Huang, Q.et al.Unified multimodal multidomain polymer representation for property prediction.npj Computational Materials11, 153 (2025). URL https://doi.org/10.1038/ s41524-025-01652-z

2025

[19] [19]

& Sun, Z.-Y

Qiu, H. & Sun, Z.-Y. On-demand reverse design of polymers with polytao.npj Computa- tional Materials10, 273 (2024). URL https://doi.org/10.1038/s41524-024-01466-5

work page doi:10.1038/s41524-024-01466-5 2024

[20] [20]

S., Xiong, W

Savit, A., Sahu, H., Shukla, S. S., Xiong, W. & Ramprasad, R.polybart: a chemical lin- guist for polymer property prediction and generative design, 12104–12119 (Association for Computational Linguistics, 2025). URL https://doi.org/10.18653/v1/2025.findings-emnlp. 647

work page doi:10.18653/v1/2025.findings-emnlp 2025

[21] [21]

Sahu, H., Xiong, W., Savit, A., Shukla, S. S. & Ramprasad, R. Polyt5: an encoder- decoder foundation chemical language model for generative polymer design.npj Artificial Intelligence2, 30 (2026). URL https://doi.org/10.1038/s44387-026-00087-1

work page doi:10.1038/s44387-026-00087-1 2026

[22] [22]

URL https: //doi.org/10.1021/acs.chemmater.1c02061

Gurnani, R.et al.polyg2g: A novel machine learning algorithm applied to the generative design of polymer dielectrics.Chemistry of Materials33, 7008–7016 (2021). URL https: //doi.org/10.1021/acs.chemmater.1c02061

work page doi:10.1021/acs.chemmater.1c02061 2021

[23] [23]

& Wang, X

Li, W., Li, Y., Lei, Q., Wang, Z. & Wang, X. PolyRL: reinforcement learning-guided polymer generation for multi-objective polymer discovery.Digital Discovery5, 266–276 (2026). URL https://doi.org/10.1039/D5DD00272A

work page doi:10.1039/d5dd00272a 2026

[24] [24]

Yue, T., Tao, L., Varshney, V. & Li, Y. Benchmarking study of deep generative models for inverse polymer design.Digital Discovery4, 910–926 (2025). URL https://doi.org/10. 1039/D4DD00395K

2025

[25] [25]

& Luo, T

Ma, R. & Luo, T. PI1M: A benchmark database for polymer informatics.Journal of Chemical Information and Modeling60, 4684–4690 (2020). URL https://doi.org/10.1021/ acs.jcim.0c00726

2020

[26] [26]

& Yamazaki, M.PoLyInfo: Polymer database for polymeric materials design, 22–29 (IEEE, 2011)

Otsuka, S., Kuwajima, I., Hosoya, J., Xu, Y. & Yamazaki, M.PoLyInfo: Polymer database for polymeric materials design, 22–29 (IEEE, 2011). URL https://doi.org/10.1109/EIDWT. 2011.13

work page doi:10.1109/eidwt 2011

[27] [27]

URL https: //doi.org/10.1038/s43018-025-00991-6

Ferber, D.et al.Development and validation of an autonomous artificial intelligence agent for clinical decision-making in oncology.Nature Cancer6, 1337–1349 (2025). URL https: //doi.org/10.1038/s43018-025-00991-6. 14

work page doi:10.1038/s43018-025-00991-6 2025