pith. sign in

arxiv: 2605.22287 · v1 · pith:6PDR6GNUnew · submitted 2026-05-21 · 💻 cs.AI

SciCore-Mol: Augmenting Large Language Models with Pluggable Molecular Cognition Modules

Yuxuan Chen , Changwei Lv , Yunduo Xiao , Zhongjing Du , Daquan Zhou , Yukun Yan , Zheni Zeng , Zhiyuan Liu This is my paper

Pith reviewed 2026-05-22 05:22 UTC · model grok-4.3

classification 💻 cs.AI
keywords molecular cognition modulespluggable LLM augmentationtopology-aware perceptionlatent diffusion generationreaction-aware reasoningchemical task performanceopen-source chemistry AIscientific discovery tools
0
0 comments X

The pith

SciCore-Mol adds three pluggable modules to large language models to process molecular topology, generation, and reactions directly.

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

The paper proposes a modular framework called SciCore-Mol that augments large language models with specialized components for handling molecular data. These components include a topology-aware perception module, a latent diffusion-based generation module, and a reaction-aware reasoning module, all connected to the LLM through learned interfaces rather than text. This setup aims to reduce the information loss that occurs when complex molecular structures are forced into linguistic descriptions. A sympathetic reader would care because it suggests that open-source systems with only 8 billion parameters can achieve performance on chemistry tasks that rivals or exceeds that of much larger proprietary models. If successful, this approach provides a way to equip AI systems with scientific expertise in a flexible, updatable manner.

Core claim

The central claim is that coupling an LLM backbone with three deeply integrated pluggable cognitive modules—a topology-aware perception module, a latent diffusion-based molecular generation module, and a reaction-aware reasoning module—through learned representation interfaces bridges the gap between discrete linguistic symbols and topological molecular or continuous reaction data, enabling richer information exchange and leading to strong performance across molecular understanding, generation, reaction prediction, and general chemistry knowledge.

What carries the argument

The three pluggable molecular cognition modules coupled to the LLM via learned representation interfaces, which allow direct handling of molecular structures and reactions to minimize semantic noise.

Load-bearing premise

The learned representation interfaces successfully couple the three modules to the LLM backbone to enable richer information exchange than text-only feedback without causing significant information loss or semantic noise.

What would settle it

A side-by-side test where removing the learned interfaces and forcing all communication through text descriptions causes the performance on chemical tasks to drop to levels comparable to standard LLMs without the modules.

Figures

Figures reproduced from arXiv: 2605.22287 by Changwei Lv, Daquan Zhou, Yukun Yan, Yunduo Xiao, Yuxuan Chen, Zheni Zeng, Zhiyuan Liu, Zhongjing Du.

Figure 1
Figure 1. Figure 1: Overview of SciCore-Mol. The GVP encoder, diffusion decoder, and reaction transformer correspond to the Topological Perception Module, Molecular Generation Module, and Reaction Sensing Module, respectively. SciCore-Mol integrates these modules with an LLM backbone to support molecular property prediction, molecule generation, synthesis prediction, retrosynthesis, yield prediction, and captioning. modules t… view at source ↗
Figure 2
Figure 2. Figure 2: (a) Inference pipeline of SciCore-Mol. The GVP encoder, Reaction Transformer, and DiT decoder implement the Topological Perception Module, Reaction Sensing Module, and Molecular Generation Module, respectively. These pluggable modules exchange information with the LLM backbone through hidden-state interfaces. (b) Progressive training pipeline, including independent component pre-training, cross-modal align… view at source ↗
Figure 3
Figure 3. Figure 3: Per-model capability radar charts across five evaluation dimensions. Raw benchmark metrics are normalized to [0, 100] via min–max scaling (Eq. 19); the normalization procedure and metric groupings are described in the Evaluation Details section. SciCore-Mol achieves the most balanced and competitive profile overall. such as BBBP, Tox21, ClinTox, HIV, BACE, and SIDER, as well as regression tasks such as ESO… view at source ↗
Figure 4
Figure 4. Figure 4: Reaction token construction in the Reaction Sensing Module. Each token combines a GVP geometry embedding, stoichiometric amount features, and a functional role signal. Masked targets and a [CLS] token enable joint product and yield prediction under a unified architecture. be present, missing values are treated as masked entries, which unifies product prediction, retrosynthesis, and yield estimation under t… view at source ↗
read the original abstract

Large Language Models (LLMs) are central to the one-for-all intelligent paradigm, but they face a fundamental challenge when dealing with heterogeneous scientific data such as molecules: the inherent gap between discrete linguistic symbols and topological molecular or continuous reaction data leads to significant information loss and semantic noise in text-based reasoning. We propose SciCore-Mol, a modular framework that bridges this gap through three deeply integrated pluggable cognitive modules: a topology-aware perception module, a latent diffusion-based molecular generation module, and a reaction-aware reasoning module. Each module is coupled to the LLM backbone through learned representation interfaces, enabling richer information exchange than is possible with text-only tool feedback. Our experiments on diverse chemical tasks demonstrate that SciCore-Mol achieves strong comprehensive performance across molecular understanding, generation, reaction prediction, and general chemistry knowledge, with an 8B-parameter open-source system that is competitive with and in several dimensions surpasses proprietary large models. This work provides a systematic blueprint for equipping LLMs with scientific expertise through decoupled, pluggable, and flexibly orchestrated modules, with direct implications for drug design, chemical synthesis, and broader scientific 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.

Referee Report

2 major / 3 minor

Summary. The manuscript introduces SciCore-Mol, a modular framework that augments LLMs with three pluggable cognitive modules—a topology-aware perception module, a latent diffusion-based molecular generation module, and a reaction-aware reasoning module—coupled to the LLM backbone via learned representation interfaces. These interfaces are intended to enable richer information exchange than text-only tool feedback, addressing the gap between discrete linguistic symbols and topological or continuous molecular data. Experiments on diverse chemical tasks are reported to demonstrate that the resulting 8B-parameter open-source system achieves strong comprehensive performance across molecular understanding, generation, reaction prediction, and general chemistry knowledge, competitive with and in several dimensions surpassing proprietary large models. The work positions itself as a blueprint for equipping LLMs with scientific expertise through decoupled, pluggable modules.

Significance. If the reported results hold, the pluggable modular design provides a systematic and extensible approach for integrating domain-specific scientific cognition into LLMs, with direct relevance to drug design, chemical synthesis, and broader scientific discovery. The open-source release of an 8B system and the emphasis on learned interfaces rather than text-only feedback represent concrete strengths that could facilitate reproducibility and further development in AI for chemistry.

major comments (2)
  1. [§4.3] §4.3 (Integration and Coupling): The claim that learned representation interfaces deliver richer exchange than text-only tool feedback is central to the framework, yet the manuscript provides only qualitative descriptions without quantitative metrics (e.g., mutual information or reconstruction error) comparing the two coupling strategies on the same tasks.
  2. [Table 2] Table 2 (Main Results): While the 8B model is stated to surpass proprietary models in several dimensions, the table lacks error bars, statistical significance tests, and explicit baseline configurations (e.g., which version of GPT-4 or Claude was used), making it difficult to assess the robustness of the competitiveness claim.
minor comments (3)
  1. [§3.2] The notation for the learned representation interfaces (e.g., the mapping functions between module outputs and LLM hidden states) is introduced without a clear equation or diagram in §3.2, which would improve clarity.
  2. [Figure 3] Figure 3 (Module Architecture) would benefit from explicit labels indicating the dimensionality of the latent spaces and the training objectives for each interface.
  3. [Related Work] A few references to prior work on molecular LLMs (e.g., in the related work section) appear to miss recent 2024 papers on similar modular approaches; adding 2–3 citations would strengthen the positioning.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback and the recommendation for minor revision. We address each major comment point by point below, indicating where revisions will be incorporated to improve clarity and rigor.

read point-by-point responses

  1. Referee: [§4.3] §4.3 (Integration and Coupling): The claim that learned representation interfaces deliver richer exchange than text-only tool feedback is central to the framework, yet the manuscript provides only qualitative descriptions without quantitative metrics (e.g., mutual information or reconstruction error) comparing the two coupling strategies on the same tasks.

    Authors: We agree that quantitative metrics would provide stronger support for the central claim. The current manuscript prioritizes end-to-end task performance to highlight practical utility, but we will revise §4.3 to include a comparative analysis subsection. This will report mutual information between learned module representations and task outcomes, along with reconstruction errors for molecular topologies under learned interfaces versus text-only baselines, derived from re-analysis of existing experimental data. revision: yes

  2. Referee: [Table 2] Table 2 (Main Results): While the 8B model is stated to surpass proprietary models in several dimensions, the table lacks error bars, statistical significance tests, and explicit baseline configurations (e.g., which version of GPT-4 or Claude was used), making it difficult to assess the robustness of the competitiveness claim.

    Authors: We acknowledge that error bars, statistical tests, and precise baseline specifications are necessary for robust evaluation. In the revised manuscript, we will update Table 2 to include error bars from repeated runs where computationally feasible, add results from statistical significance tests (e.g., paired t-tests or Wilcoxon tests), and explicitly document the exact model versions (such as GPT-4-0613 and Claude 3 Opus) along with prompting details. The experimental setup section will be expanded accordingly. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper introduces a modular framework (SciCore-Mol) consisting of three pluggable cognitive modules coupled to an LLM backbone via learned representation interfaces. All central claims rest on empirical experiments across molecular understanding, generation, reaction prediction, and chemistry knowledge tasks, with performance reported for an 8B open-source system. No equations, derivations, fitted parameters, self-definitional constructions, or load-bearing self-citations appear in the abstract or described architecture. The integration premise is presented as directly supported by the reported results rather than reducing to its own inputs by construction. The work is therefore self-contained against external benchmarks with no circular reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are identifiable from the abstract alone; the framework description does not specify any fitted constants or new postulated entities.

pith-pipeline@v0.9.0 · 5754 in / 1053 out tokens · 31597 ms · 2026-05-22T05:22:33.610626+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

58 extracted references · 58 canonical work pages · 8 internal anchors

  1. [1]

    ChemBERTa- 2: Towards chemical foundation models.arXiv preprint arXiv:2209.01712, 2022

    Walid Ahmad, Elana Simon, Seyone Chithrananda, Gabriel Grand, and Bharath Ramsundar. ChemBERTa- 2: Towards chemical foundation models.arXiv preprint arXiv:2209.01712, 2022. URLhttps: //arxiv.org/abs/2209.01712

    work page arXiv 2022

  2. [2]

    Intern-s1: A scientific multimodal foundation model.arXiv preprint arXiv:2508.15763, 2025

    Lei Bai, Zhongrui Cai, Yuhang Cao, Maosong Cao, Weihan Cao, Chiyu Chen, Haojiong Chen, Kai Chen, Pengcheng 12 Chen, Ying Chen, et al. Intern-s1: A scientific multimodal foundation model.arXiv preprint arXiv:2508.15763, 2025

    work page arXiv 2025

  3. [3]

    Meteor: An automatic metric for mt evaluation with improved correlation with human judgments

    Satanjeev Banerjee and Alon Lavie. Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. InProceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, pages 65–72, 2005

    work page 2005

  4. [4]

    Boiko, Robert MacKnight, and Gabe Gomes

    David A. Boiko, Robert MacKnight, and Gabe Gomes. Autonomous chemical research with large language models. Nature, 2023. URLhttps://www.nature.com/ articles/s41586-023-06792-0

    work page 2023

  5. [5]

    Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D

    Andres M. Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D. White, and Philippe Schwaller. Augmenting large language models with chemistry tools.Nature Chemistry,

    work page

  6. [6]

    URLhttps://arxiv.org/abs/2304.05376

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    LDMol: Text-to-molecule diffusion model with structurally informative latent space

    Jinho Chang and Jong Chul Ye. LDMol: Text-to-molecule diffusion model with structurally informative latent space. arXiv preprint arXiv:2405.17829, 2024. URLhttps:// arxiv.org/abs/2405.17829

    work page arXiv 2024

  8. [8]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. InInternational Conference on Machine Learning, pages 1597–1607, 2020

    work page 2020

  9. [9]

    Translation between Molecules and Natural Language

    Carl Edwards, Tuan Lai, Kevin Ros, Garrett Honke, Kyunghyun Cho, and Heng Ji. Translation between molecules and natural language.arXiv preprint arXiv:2204.11817, 2022. URLhttps://arxiv.org/abs/2204.11817

    work page arXiv 2022

  10. [10]

    Grzybowski, Ying Diao, Jiawei Han, Ge Liu, Hao Peng, Martin D

    Carl Edwards, Chi Han, Gawon Lee, Thao Nguyen, Bowen Jin, Chetan Kumar Prasad, Sara Szymku ´c, Bartosz A. Grzybowski, Ying Diao, Jiawei Han, Ge Liu, Hao Peng, Martin D. Burke, and Heng Ji. mclm: A function-infused and synthesis-friendly modular chemical language model.arXiv preprint arXiv:2505.12565, 2025. URLhttps://arxiv. org/abs/2505.12565

    work page arXiv 2025

  11. [12]

    URLhttps://arxiv.org/abs/2306.08018

    work page arXiv

  12. [13]

    Measuring Massive Multitask Language Understanding

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding.arXiv preprint arXiv:2009.03300, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  13. [14]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. InAdvances in Neural Information Processing Systems, 2020

    work page 2020

  14. [15]

    Equivariant diffusion for molecule generation in 3d

    Emiel Hoogeboom, V ´ıctor Garc´ıa Satorras, Cl´ement Vignac, and Max Welling. Equivariant diffusion for molecule generation in 3d. InProceedings of the 39th International Conference on Machine Learning, 2022

    work page 2022

  15. [16]

    GPT-4o System Card

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card.arXiv preprint arXiv:2410.21276, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  16. [17]

    Chemformer: A pre-trained transformer for computational chemistry.Machine Learning: Science and Technology, 3(1):015022, 2022

    Ross Irwin, Spyridon Dimitriadis, Jiazhen He, and Esben Jan- nik Bjerrum. Chemformer: A pre-trained transformer for computational chemistry.Machine Learning: Science and Technology, 3(1):015022, 2022

    work page 2022

  17. [18]

    Cumulated gain- based evaluation of ir techniques.ACM Transactions on Information Systems (TOIS), 20(4):422–446, 2002

    Kalervo J ¨arvelin and Jaana Kek ¨al¨ainen. Cumulated gain- based evaluation of ir techniques.ACM Transactions on Information Systems (TOIS), 20(4):422–446, 2002

    work page 2002

  18. [19]

    Junction tree variational autoencoder for molecular graph generation

    Wengong Jin, Regina Barzilay, and Tommi Jaakkola. Junction tree variational autoencoder for molecular graph generation. InProceedings of the 35th International Conference on Machine Learning, pages 2323–2332, 2018

    work page 2018

  19. [20]

    Bowen Jing, Stephan Eismann, Patricia Suriana, Raphael J. L. Townshend, and Ron Dror. Learning from protein structure with geometric vector perceptrons. InInternational Conference on Learning Representations, 2021

    work page 2021

  20. [21]

    The open reaction database.Journal of the American Chemical Society, 143(45):18820–18826, 2021

    Steven M Kearnes, Michael R Maser, Michael Wleklinski, Anton Kast, Abigail G Doyle, Spencer D Dreher, Joel M Hawkins, Klavs F Jensen, and Connor W Coley. The open reaction database.Journal of the American Chemical Society, 143(45):18820–18826, 2021

    work page 2021

  21. [22]

    Semi-Supervised Classification with Graph Convolutional Networks

    Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks.arXiv preprint arXiv:1609.02907, 2016. URLhttps://arxiv. org/abs/1609.02907

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  22. [23]

    Directional message passing for molecular graphs

    Johannes Klicpera, Janek Groß, and Stephan G ¨unnemann. Directional message passing for molecular graphs. In International Conference on Learning Representations, 2020

    work page 2020

  23. [24]

    SELFIES and the future of molecular string representations.Patterns, 1(9):100099, 2020

    Mario Krenn, Florian H ¨ase, AkshatKumar Nigam, Pascal Friederich, and Al´an Aspuru-Guzik. SELFIES and the future of molecular string representations.Patterns, 1(9):100099, 2020

    work page 2020

  24. [25]

    Rdkit documentation.Release, 1(1-79): 4, 2013

    Greg Landrum et al. Rdkit documentation.Release, 1(1-79): 4, 2013

    work page 2013

  25. [26]

    Towards 3d molecule-text interpretation in language models.arXiv preprint arXiv:2401.13923, 2024

    Sihang Li, Zhiyuan Liu, Yanchen Luo, Xiang Wang, Xiangnan He, Kenji Kawaguchi, Tat-Seng Chua, and Qi Tian. Towards 3d molecule-text interpretation in language models.arXiv preprint arXiv:2401.13923, 2024. URLhttps://arxiv. org/abs/2401.13923

    work page arXiv 2024

  26. [27]

    Drugr: Optimizing molecular drugs through llm-based explicit reasoning.arXiv preprint arXiv:2602.08213, 2026

    Haoran Liu, Zheni Zeng, Yukun Yan, Yuxuan Chen, and Yunduo Xiao. Drugr: Optimizing molecular drugs through llm-based explicit reasoning.arXiv preprint arXiv:2602.08213, 2026

    work page arXiv 2026

  27. [28]

    GIT- Mol: A multi-modal large language model for molecular science with graph, image, and text.arXiv preprint arXiv:2308.06911, 2023

    Pengfei Liu, Yiming Ren, Jun Tao, and Zhixiang Ren. GIT- Mol: A multi-modal large language model for molecular science with graph, image, and text.arXiv preprint arXiv:2308.06911, 2023. URLhttps://arxiv.org/ abs/2308.06911

    work page arXiv 2023

  28. [29]

    Multi-modal molecule structure– text model for text-based retrieval and editing.Nature Machine Intelligence, 5(12):1447–1457, 2023

    Shengchao Liu, Weili Nie, Chengpeng Wang, Jiarui Lu, Zhuoran Qiao, Ling Liu, Jian Tang, Chaowei Xiao, and Animashree Anandkumar. Multi-modal molecule structure– text model for text-based retrieval and editing.Nature Machine Intelligence, 5(12):1447–1457, 2023. doi: 10.1038/ s42256-023-00759-6

    work page 2023

  29. [30]

    Molca: Molecular graph-language modeling with cross-modal projector and uni-modal adapter

    Zhiyuan Liu, Sihang Li, Yanchen Luo, Hao Fei, Yixin Cao, Kenji Kawaguchi, Xiang Wang, and Tat-Seng Chua. Molca: Molecular graph-language modeling with cross-modal projector and uni-modal adapter. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, 2024

    work page 2024

  30. [31]

    Molfm: A multimodal molecular foundation model.arXiv preprint arXiv:2307.09484, 2023

    Yizhen Luo, Kai Yang, Massimo Hong, Xing Yi Liu, and Zaiqing Nie. Molfm: A multimodal molecular foundation model.arXiv preprint arXiv:2307.09484, 2023

    work page arXiv 2023

  31. [32]

    A framework for evaluating the chemical knowledge and reasoning abilities of large language models against the expertise of chemists.Nature Chemistry, 17:1027–1034, 13

    Adrian Mirza, Nawaf Alampara, Sreekanth Kunchapu, et al. A framework for evaluating the chemical knowledge and reasoning abilities of large language models against the expertise of chemists.Nature Chemistry, 17:1027–1034, 13

    work page

  32. [33]

    doi: 10.1038/s41557-025-01815-x

    work page doi:10.1038/s41557-025-01815-x

  33. [34]

    Nemotron-cc: Transforming common crawl into a refined long-horizon pretraining dataset, 2024

    NVIDIA. Nemotron-cc: Transforming common crawl into a refined long-horizon pretraining dataset, 2024. URL https://huggingface.co/datasets/nvidia/ Nemotron-CC

    work page 2024

  34. [35]

    SMolInstruct.https: //huggingface.co/datasets/osunlp/ SMolInstruct, 2024

    OSU NLP Group. SMolInstruct.https: //huggingface.co/datasets/osunlp/ SMolInstruct, 2024. Large-scale chemistry instruction- tuning dataset

    work page 2024

  35. [36]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 4195– 4205, 2023

    work page 2023

  36. [37]

    BioT5: Enriching Cross-modal Integration in Biology with Chemical Knowledge and Natural Language Associations

    Qizhi Pei, Wei Zhang, Jinhua Zhu, Kehan Wu, Kaiyuan Gao, Lijun Wu, Yingce Xia, and Rui Yan. BioT5: Enriching cross-modal integration in biology with chemical knowledge and natural language associations.arXiv preprint arXiv:2310.07276, 2023. URLhttps://arxiv.org/ abs/2310.07276

    work page arXiv 2023

  37. [38]

    E(n) equivariant graph neural networks

    V ´ıctor Garc´ıa Satorras, Emiel Hoogeboom, and Max Welling. E(n) equivariant graph neural networks. InProceedings of the 38th International Conference on Machine Learning, pages 9323–9332, 2021

    work page 2021

  38. [39]

    Sch ¨utt, Huziel E

    Kristof T. Sch ¨utt, Huziel E. Sauceda, Pieter-Jan Kindermans, Alexandre Tkatchenko, and Klaus-Robert M ¨uller. SchNet: A deep learning architecture for molecules and materials.The Journal of Chemical Physics, 148(24):241722, 2018

    work page 2018

  39. [40]

    Sch ¨utt, Oliver T

    Kristof T. Sch ¨utt, Oliver T. Unke, and Michael Gastegger. Equivariant message passing for the prediction of tensorial properties and molecular spectra. InProceedings of the 38th International Conference on Machine Learning, pages 9377– 9388, 2021

    work page 2021

  40. [41]

    OpenAI GPT-5 System Card

    Aaditya Singh, Adam Fry, Adam Perelman, Adam Tart, Adi Ganesh, Ahmed El-Kishky, Aidan McLaughlin, Aiden Low, AJ Ostrow, Akhila Ananthram, et al. Openai gpt-5 system card.arXiv preprint arXiv:2601.03267, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  41. [42]

    ChemAgent: Self-updating memories in large language models improves chemical reasoning

    Xiangru Tang, Tianyu Hu, Muyang Ye, Yanjun Shao, Xunjian Yin, Siru Ouyang, Wangchunshu Zhou, Pan Lu, Zhuosheng Zhang, Yilun Zhao, Arman Cohan, and Mark Gerstein. ChemAgent: Self-updating memories in large language models improves chemical reasoning. InInternational Conference on Learning Representations, 2025. URL https://arxiv.org/abs/2501.06590

    work page arXiv 2025

  42. [43]

    Mahoney, Andy Nonaka, and Zhi Yao

    Yingheng Tang, Wenbin Xu, Jie Cao, Weilu Gao, Steven Farrell, Benjamin Erichson, Michael W. Mahoney, Andy Nonaka, and Zhi Yao. MatterChat: A multi-modal LLM for material science.arXiv preprint arXiv:2502.13107, 2025. URLhttps://arxiv.org/abs/2502.13107

    work page arXiv 2025

  43. [45]

    URLhttps://arxiv.org/abs/2211.09085

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [46]

    DiGress: Discrete denoising diffusion for graph generation

    Cl ´ement Vignac, Igor Krawczuk, Antoine Siraudin, Bohan Wang, V olkan Cevher, and Pascal Frossard. DiGress: Discrete denoising diffusion for graph generation. InInternational Conference on Learning Representations, 2023

    work page 2023

  45. [47]

    MiDi: Mixed graph and 3d denoising diffusion for molecule generation

    Cl ´ement Vignac, Nagham Osman, Laura Toni, and Pascal Frossard. MiDi: Mixed graph and 3d denoising diffusion for molecule generation. InJoint European Conference on Machine Learning and Knowledge Discovery in Databases, 2023

    work page 2023

  46. [48]

    SMILES-BERT: Large scale unsupervised pre- training for molecular property prediction

    Sheng Wang, Yuzhi Guo, Yifei Wang, Hao Sun, and Junzhou Huang. SMILES-BERT: Large scale unsupervised pre- training for molecular property prediction. InProceedings of the 10th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics, pages 429– 436, 2019

    work page 2019

  47. [49]

    Molecular contrastive learning of representations via graph neural networks.Nature Machine Intelligence, 4(3):279–287, 2022

    Yuyang Wang, Jianren Wang, Zhonglin Cao, and Amir Barati Farimani. Molecular contrastive learning of representations via graph neural networks.Nature Machine Intelligence, 4(3):279–287, 2022

    work page 2022

  48. [50]

    Moleculenet: a benchmark for molecular machine learning.Chemical science, 9(2):513–530, 2018

    Zhenqin Wu, Bharath Ramsundar, Evan N Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S Pappu, Karl Leswing, and Vijay Pande. Moleculenet: a benchmark for molecular machine learning.Chemical science, 9(2):513–530, 2018

    work page 2018

  49. [51]

    How Powerful are Graph Neural Networks?

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks?arXiv preprint arXiv:1810.00826, 2018. URLhttps://arxiv.org/ abs/1810.00826

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  50. [52]

    GeoDiff: A geometric diffusion model for molecular conformation generation

    Minkai Xu, Lantao Yu, Yang Song, Chence Shi, Stefano Ermon, and Jian Tang. GeoDiff: A geometric diffusion model for molecular conformation generation. InInternational Conference on Learning Representations, 2022

    work page 2022

  51. [53]

    Geometric latent diffusion models for 3d molecule generation

    Minkai Xu, Alexander Powers, Ron Dror, and Stefano Ermon. Geometric latent diffusion models for 3d molecule generation. InProceedings of the 40th International Conference on Machine Learning, 2023

    work page 2023

  52. [54]

    Qwen3 Technical Report

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  53. [55]

    Do transformers really perform bad for graph representation? In Advances in Neural Information Processing Systems, 2021

    Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, and Tie-Yan Liu. Do transformers really perform bad for graph representation? In Advances in Neural Information Processing Systems, 2021

    work page 2021

  54. [57]

    URLhttps://arxiv.org/abs/2402.09391

    work page arXiv

  55. [58]

    MoFlow: An invertible flow model for generating molecular graphs

    Chengxi Zang and Fei Wang. MoFlow: An invertible flow model for generating molecular graphs. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 617–626, 2020

    work page 2020

  56. [59]

    ChatMol: Interactive molecular discovery with natural language.Bioinformatics, 40(9): btae534, 2024

    Zheni Zeng, Bangchen Yin, Shipeng Wang, Jiarui Liu, Cheng Yang, Haishen Yao, Xingzhi Sun, Maosong Sun, Guotong Xie, and Zhiyuan Liu. ChatMol: Interactive molecular discovery with natural language.Bioinformatics, 40(9): btae534, 2024. doi: 10.1093/bioinformatics/btae534

    work page doi:10.1093/bioinformatics/btae534 2024

  57. [60]

    Chemllm: A chemical large language model

    Di Zhang, Wei Liu, Qian Tan, Jingdan Chen, Hang Yan, Yuliang Yan, Jiatong Li, Weiran Huang, Xiangyu Yue, Wanli Ouyang, Dongzhan Zhou, Shufei Zhang, Mao Su, Han-Sen Zhong, and Yuqiang Li. ChemLLM: A chemical large language model.arXiv preprint arXiv:2402.06852, 2024. URLhttps://arxiv.org/abs/2402.06852

    work page arXiv 2024

  58. [61]

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

    Gengmo Zhou, Zhifeng Gao, Qiankun Ding, Hang Zheng, Hongteng Xu, Zhewei Wei, Linfeng Zhang, and Guolin Ke. Uni-Mol: A universal 3d molecular representation learning framework. InInternational Conference on Learning 14 Representations, 2023. 15

    work page 2023