Pith. sign in

REVIEW 3 major objections 6 minor 2 cited by

Mol-LLM: Multimodal Generalist Molecular LLM with Improved Graph Utilization

T0 review · 3 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that a single multimodal molecular LLM, Mol-LLM, outperforms prior generalist molecular LLMs across property, reaction, generation, and captioning tasks by training it to prefer correct molecular graphs over perturbed ones.

desk verdict Mol-LLM is a real empirical advance for generalist molecular LLMs, but the graph-utilization evidence is circular and the reporting needs cleanup. read the letter →

arxiv 2502.02810 v2 pith:VLMG63SJ submitted 2025-02-05 cs.LG cs.AIphysics.chem-phq-bio.BM

Chanhui Lee , Hanbum Ko , Yuheon Song , YongJun Jeong , Rodrigo Hormazabal , Sehui Han , Kyunghoon Bae , Sungbin Lim
show 1 more author
Sungwoong Kim
This is my paper · ORCID
classification cs.LGcs.AIphysics.chem-phq-bio.BM
keywords molecularlanguagemodelmultimodallearninggraphrepresentationpreferenceoptimizationinstructiontuningpropertypredictionreactionout-of-distributiongeneralization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that a molecular large language model can be a true generalist—handling property regression, property classification, reaction prediction, molecule generation, and captioning in one model—if it is explicitly trained to use 2D molecular graphs rather than only SMILES-like strings. The authors diagnose a graph-bypass problem: naive next-token training leaves the model unable to tell a molecule apart from a perturbed version of it, so they propose Molecular structure Preference Optimization (MolPO), which trains on preferences between correct and perturbed graphs, together with a hybrid graph encoder pre-trained to predict functional groups and reconstruct SELFIES. If the central claim holds, a single instruction-tuned model can replace collections of task-specific chemistry models and generalize to new distributions better than sequence-only generalists.

What carries the argument

The load-bearing mechanism is the MolPO objective, which defines per-task rewards $r_{w,i}$ and $r_{\ell,i}$ from the average log-likelihood of the answer under the chosen and rejected graphs, and optimizes $-\log\sigma(\min(r_{w,i}-r_{\ell,i},\lambda_{\mathrm{clip}}|r_{w,i}|)-\gamma_i)$ with a task-adaptive target reward margin $\gamma_i$. Rejected graphs are produced by randomly removing and adding MACCS-key substructures, so the preference pairs require no task-specific perturbation design. Supporting this are a hybrid graph encoder that concatenates GINE local message-passing embeddings with TokenGT global transformer embeddings (node, edge, and graph tokens) for the Q-Former cross-modal projector, and a two-part GNN pre-training objective—functional group prediction plus SELFIES reconstruction—that teaches the encoder which substructures matter and how to preserve global structure.

What would settle it

Build a test where the graph perturbation removes or adds a substructure known to be irrelevant to the target property, and check whether the model still assigns higher likelihood to the original graph; if the graph discrimination ratio stays near 1 on such property-preserving edits, the preference signal is responding to edit artifacts, not to chemistry.

Watch

Extended reading notes

Core claim

The central claim is that the graph-bypass problem—an LLM trained with supervised fine-tuning assigns nearly equal likelihood to a molecule and its perturbed graph—can be fixed by preference optimization over molecular structures, and that this fix is what lets a multimodal generalist outperform sequence-only generalists. The paper reports that Mol-LLM achieves state-of-the-art or comparable results on a benchmark spanning all five task groups and beats prior generalist molecular LLMs by a large margin on out-of-distribution reaction and property prediction. A graph-discrimination ratio confirms that the model now ranks the correct graph above the perturbed one on most tasks, which the authors take as evidence that the graph modality is actually being used.

Load-bearing premise

MolPO assumes that randomly removing and adding MACCS-key substructures produces rejected graphs that are genuinely worse for the target task, so the preference signal teaches task-relevant chemistry rather than artifacts of the editing procedure.

Editorial extensions

If this is right

  • On most benchmark tasks, one Mol-LLM model matches or beats specialist, semi-generalist, and generalist baselines, so a single generalist can cover the five molecular task groups.
  • The largest gains appear on out-of-distribution reaction and solubility benchmarks, which suggests graph-aware preference training transfers to distributions not seen during instruction tuning.
  • The graph discrimination ratio provides a label-free check that a multimodal molecular model is actually using the graph modality, and it can be measured per task.
  • On the small property-classification datasets the MolPO gain largely disappears, indicating that preference optimization needs diverse molecular distributions to help.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the perturbation recipe is task-agnostic and cheap, the same MolPO idea could be applied to other multimodal molecular LLMs or to 3D conformer inputs, a direction the authors only list as future work.
  • A stress test with property-preserving edits would reveal whether MolPO learns chemistry or simply detects the presence of an edit; the paper does not run this control.
  • The graph discrimination ratio could serve as a general diagnostic for graph-conditioned LLMs, letting practitioners detect graph bypass before spending compute on full fine-tuning.
  • If the OOD gains hold beyond the two constructed benchmarks, preference-based graph grounding may be a cheaper route to distributional robustness than scaling sequence-only instruction data.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes Mol-LLM, a multimodal generalist molecular LLM that combines a hybrid graph encoder (GINE + TokenGT), a Q-Former cross-modal projector, and a Mistral-7B backbone. The two main technical contributions are (i) Molecular structure Preference Optimization (MolPO), which trains the model to prefer an original molecular graph over a MACCS-key-perturbed graph, and (ii) a GNN pre-training scheme with functional-group prediction and SELFIES reconstruction. The authors construct a ~3.3M-example instruction-tuning dataset and two OOD benchmarks (AqSol for LogS, ORDerly for reaction prediction). The experiments report that Mol-LLM achieves state-of-the-art or comparable results among generalist molecular LLMs across property regression, property classification, reaction prediction, molecule generation, and molecule captioning, with particularly large OOD gains in reaction prediction.

Significance. If the claims hold, the paper is a useful empirical contribution: it demonstrates a single model covering five task groups, introduces OOD benchmarks that the community can reuse, and provides a concrete recipe (graph encoder pre-training + preference optimization) that may help future multimodal molecular LLMs. The paper is transparent about hyperparameters, training resources, and limitations, and it states that code, data, and checkpoints will be released. However, the central evidence for the paper's key novelty—'improved graph utilization'—is not conclusive: the graph discrimination ratio (GDR) is computed on the same perturbation distribution used to train MolPO, so high GDR may reflect successful fitting to a specific perturbation artifact rather than genuine chemical graph understanding. The downstream performance gains attributable to MolPO are small (Table 5) and are reported without error bars, so the significance of the headline claim is currently limited.

major comments (3)
  1. [§3.3, Eq. GDR and Appendix B.4] The graph discrimination ratio (GDR) defined in Section 3.3 is circular as evidence of graph utilization. The rejected graphs gℓ are generated by the same random MACCS-key removal/addition procedure described in Appendix B.4 that is used to construct MolPO training pairs. Since the MolPO objective in Eq. (1) explicitly increases rw,i relative to rℓ,i on this exact distribution, a high GDR on this distribution is largely a train-set reproduction effect; it does not demonstrate that the model has learned task-relevant chemical structure rather than perturbation-detection artifacts. The authors should either evaluate GDR on held-out perturbation schemes (e.g., different modification ratios, alternative substructure perturbations, or chemically meaningful mutations) or provide independent downstream evidence that MolPO improves genuine graph understanding.
  2. [Appendix A] The limitation section explicitly acknowledges that 'if the training molecular distribution is too narrow or contains spurious patterns unrelated to the given molecular task, the model may inappropriately regard molecules in test set or OOD dataset as rejected molecules,' yet no experiment tests whether this actually occurs. This is a load-bearing concern because the MolPO hypothesis is precisely that the MACCS-perturbed graphs are task-relevant negatives. The paper should include a direct test, for example, applying MolPO with perturbation edits that preserve the target property and measuring whether the model still learns a preference, or measuring whether GDR drops when the perturbation distribution is shifted.
  3. [Checklist item 7 and Tables 1–5] The paper reports no error bars or statistical significance tests (its own checklist answers 'No'), and many of the claimed improvements are very small: for example, Table 5 shows FS (SMol.) 0.598 to 0.601, T2M (ChEBI-20) 0.426 to 0.443, and LogD 0.96 to 0.91, while Table 1 shows Mol-LLM underperforming LlaSMol on SIDER (76.3 vs 78.4) and BBBP (81.1 vs 82.4). Without multiple seeds or confidence intervals, the key claim that MolPO improves performance on 'most tasks' is not statistically substantiated. The authors should report error bars over at least 3 runs for the ablation study and for the main competitive comparisons where margins are small.
minor comments (6)
  1. [§3.2, paragraph on property tasks] The text states that 'even Mol-LLM (w/o Graph) performs on a par with the full model' on MoleculeNet tasks and attributes this to small molecular sizes; this observation undercuts the paper's emphasis on graph utilization as a general contribution and should be discussed in the introduction or conclusion.
  2. [§3.3, Table 3] For molecule captioning, Mol-LLM shows lower ROUGE-L than the w/o Graph variant on both ChEBI-20 and SMolInstruct, and the explanation that it 'produces more concise captions' is post hoc; please provide supporting evidence or temper the claim.
  3. [§1, Introduction] The novelty claim that Mol-LLM is 'the first versatile generalist multimodal molecular LLM' should be qualified, since related work UniMoT and 3D-MolT5 also combine structure encoders with instruction tuning; please clarify what distinguishes Mol-LLM from these models.
  4. [§2.2, Eq. (1)] In the MolPO objective, the margin clipping term min(rw,i − rℓ,i, λclip|rw,i|) may become negative if rw,i is negative; please clarify the intended behavior of the clipping for negative chosen rewards.
  5. [Appendix B.4 and Table 7] The hyperparameters λmargin = 0.5, λclip = 1.0, c = 0.25, and the MACCS perturbation ratio (30%) are not ablated; a sensitivity study would help establish the robustness of the proposed method.
  6. [Throughout] There are several typos and formatting issues, including 'bencharks' in the introduction, 'V ALIDITY' in Table 3, 'Heat capavity' in Table 6, 'ROGUE' in Table 13, and inconsistent spacing in Tables 1–4; these should be corrected.

Circularity Check

1 steps flagged · score 6.0 of 10

Graph-utilization evidence is circular: GDR counts the same chosen-vs-rejected reward margin that MolPO optimizes, on the same MACCS perturbation distribution.

  1. fitted input called prediction [Section 2.2 Stage 3, Eq. (1); Section 3.3; Appendix B.4]
    "LMolPO = E(s,qi,g,y)∼Dtr [− log σ(min(rw,i − rℓ,i, λclip|rw,i|) − γi)] ... We then compute the graph discrimination ratio GDR = 1/Ni Σ_n I(rw,i(n) > rℓ,i(n)) ... The consistently higher GDRs in the MolPO setting confirm that this objective helps the model make better use of molecular graph information. ... we employed functional group-based substructure modification ... MACCS keys-based substructure modification method directly modifies molecular substructures by randomly removing and adding them."

    GDR counts the fraction of instances where the chosen-graph reward rw exceeds the rejected-graph reward rℓ, which is exactly the preference margin that LMolPO is trained to increase: the objective minimizes -log σ(min(rw − rℓ, λclip|rw|) − γi), pushing rw above rℓ by at least the task-adaptive margin. A high GDR on chosen-vs-rejected pairs is therefore the training objective's success criterion, not an independent measurement of chemical graph understanding. The rejected graphs are generated by the same MACCS-key random removal/addition procedure used to construct MolPO training pairs (Appendix B.4), so the evaluation distribution coincides with the trained perturbation distribution.

full rationale

The main benchmark comparisons (Tables 1-4) and the GNN-pretraining experiment (Figure 5) are independent of the MolPO preference objective and are not circular: they use external task labels and compare against baselines. The single load-bearing circular step is the graph-utilization evidence in Figure 1/Section 3.3. GDR is defined as the frequency with which the chosen-graph reward exceeds the rejected-graph reward, and LMolPO (Eq. 1) is explicitly optimized to create that margin on pairs from the same MACCS-key perturbation distribution (Appendix B.4). Consequently, the high GDR of the MolPO-trained model mostly reflects successful optimization of the training objective, and Appendix A's limitation passage concedes that the model may rely on spurious distributional patterns rather than task-relevant chemistry. This does not collapse the paper's overall contribution, because the downstream performance gains and OOD results are measured on real task labels; but the specific claim of 'demonstrably better graph utilization' rests on a metric that is by construction tied to the training target. Score 6 reflects this partial, localized circularity.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim rests on empirical training and evaluation rather than a formal derivation. The load-bearing choices are the preference pair construction, the pretraining objectives, the hybrid encoder architecture, and the scale of instruction data. No new physical entities are introduced.

free parameters (7)
  • lambda_margin = 0.5 (B.5) or 0.25 (Table 7)
    Task-adaptive target reward margin scale in MolPO; reported inconsistently across the paper.
  • lambda_clip = 1.0
    Margin clipping coefficient in the MolPO objective, chosen by hand.
  • c = 0.25
    Scaling factor balancing SFT and MolPO losses.
  • num_query_tokens = 32
    Number of Q-Former query tokens, chosen for multitask learning and larger than prior work.
  • MACCS_perturbation_ratio = 0.3
    Fraction of present MACCS keys randomly removed and absent keys added to build rejected graphs; arbitrary choice.
  • K_functional_groups = 72
    Number of functional groups retained for pretraining after discarding 11 most common and 1 rarest group.
  • GNN_pretraining_epochs = 50
    Number of epochs for GINE and TokenGT pretraining.
assumptions (5)
  • domain assumption 2D molecular graphs contain task-relevant structural information beyond 1D SELFIES strings.
    Motivates the graph encoder and MolPO; invoked throughout Sections 1 and 2.
  • ad hoc to paper Randomly removing and adding MACCS-key substructures produces rejected graphs whose target labels differ from the original molecule.
    Preference pair construction in Appendix B.4 assumes perturbation alters task-relevant features without task-specific design.
  • domain assumption Large-scale instruction tuning on 3.3M examples transfers to out-of-distribution tasks.
    OOD evaluation with AqSol and ORDerly assumes the training distribution supports generalization to new molecule and reaction distributions.
  • domain assumption Functional group prediction and SELFIES reconstruction pretraining improve downstream molecular representations.
    Stage 1 pretraining objectives in Section 2.2 assume these self-supervised signals transfer to downstream tasks.
  • domain assumption Mistral-7B backbone with Q-Former alignment has sufficient capacity and inductive bias for molecular tasks.
    Architecture choice inherited from prior molecular LLMs and vision-language models.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mol-LLM: Multimodal Generalist Molecular LLM with Improved Graph Utilization." pith.science (2026). https://pith.science/paper/VLMG63SJ

@misc{pith2026250202810,
  author       = {Pith},
  title        = {Pith review of: Mol-LLM: Multimodal Generalist Molecular LLM with Improved Graph Utilization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VLMG63SJ}},
  note         = {Machine review of arXiv:2502.02810}
}
read the original abstract

Recent advances in large language models (LLMs) have led to models that tackle diverse molecular tasks, such as chemical reaction prediction and molecular property prediction. Large-scale molecular instruction-tuning datasets have enabled sequence-only (e.g., SMILES or SELFIES) generalist molecular LLMs, and researchers are now exploring multimodal approaches that incorporate molecular structural information for further gains. However, a genuinely multimodal, generalist LLM that covers a broad spectrum of molecular tasks has yet to be fully investigated. We observe that naive next token prediction training ignores graph-structural information, limiting an LLM's ability to exploit molecular graphs. To address this, we propose (i) Molecular structure Preference Optimization (MolPO), which facilitates graph usage by optimizing preferences between pairs of correct and perturbed molecular structures, and (ii) an advanced graph encoder with a tailored pre-training strategy to improve the effect of graph utilization by MolPO. Building on these contributions, we introduce Mol-LLM, the first multimodal generalist model that (a) handles a broad spectrum of molecular tasks among molecular LLMs, (b) explicitly leverages molecular-structure information, and (c) takes advantage of extensive instruction tuning. Mol-LLM attains state-of-the-art or comparable results across the most comprehensive molecular-LLM benchmark-even on out-of-distribution datasets for reaction and property prediction, where it surpasses prior generalist molecular LLMs by a large margin.

Figures

Figures reproduced from arXiv: 2502.02810 by the authors.

Figure 1
Figure 1. (Left) Performance comparison among generalist molecular LLMs with normalized primary [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. (Left) Overall structure of Mol-LLM. Molecular graph is encoded into a fixed-length token [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of the three training stages and the loss function used at each stage. The training [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Comparison of fine-tuning performances on three tasks under different GNN architectures [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: (Top) Distribution of functional groups present in molecules from the PubChem database. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Comparison of predicted outputs by generalists on forward synthesis (FS) and retrosynthesis [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Breaking Bad Molecules: Are MLLMs Ready for Structure-Level Molecular Detoxification?

    cs.AI 2025-06 conditional novelty 7.0 of 10

    A new benchmark called ToxiMol evaluates how well 43 multimodal LLMs can edit toxic molecules into structurally similar, non-toxic, drug-like candidates; the best model succeeds on 43.3% of tasks.

  2. DrugGen 2: A disease-aware language model for enhancing drug discovery

    q-bio.QM 2026-07 conditional novelty 6.0 of 10

    A GPT-2 model fine-tuned with disease MeSH + protein sequence inputs and GRPO rewards produces more unique, valid, drug-like, high-PLAPT-affinity ligands than DrugGPT or DrugGen on five diabetic-nephropathy targets.

Reference graph

Works this paper leans on

73 extracted references · 38 canonical work pages · cited by 2 Pith papers

  1. [1]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-V oss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwi...

    work page 2023

  2. [2]

    Gpt-4 technical report

    OpenAI. Gpt-4 technical report. ArXiv, abs/2303.08774, 2023

    arXiv 2023

  3. [3]

    Gemini: A family of highly capable multimodal models

    Gemini Team Google. Gemini: A family of highly capable multimodal models. ArXiv, abs/2312.11805, 2023

    arXiv 2023

  4. [4]

    Llama 2: Open foundation and fine-tuned chat models, 2023

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Harts...

    arXiv 2023

  5. [5]

    Baker, Ziqi Chen, Xia Ning, and Huan Sun

    Botao Yu, Frazier N. Baker, Ziqi Chen, Xia Ning, and Huan Sun. Llasmol: Advancing large language models for chemistry with a large-scale, comprehensive, high-quality instruction tuning dataset. ArXiv, abs/2402.09391, 2024. URL https://api.semanticscholar.org/ CorpusID:267657622. 10

    arXiv 2024

  6. [6]

    Biot5+: Towards generalized biological understanding with iupac integration and multi-task tuning

    Qizhi Pei, Lijun Wu, Kaiyuan Gao, Xiaozhuan Liang, Yin Fang, Jinhua Zhu, Shufang Xie, Tao Qin, and Rui Yan. Biot5+: Towards generalized biological understanding with iupac integration and multi-task tuning. ArXiv, abs/2402.17810, 2024. URL https://api.semanticscholar. org/CorpusID:268041632

    arXiv 2024

  7. [7]

    Mol-instructions: A large-scale biomolecular instruction dataset for large language models

    Yin Fang, Xiaozhuan Liang, Ningyu Zhang, Kangwei Liu, Rui Huang, Zhuo Chen, Xiaohui Fan, and Huajun Chen. Mol-instructions: A large-scale biomolecular instruction dataset for large language models. ArXiv, abs/2306.08018, 2023. URL https://api.semanticscholar. org/CorpusID:259164901

    arXiv 2023

  8. [8]

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

    Zhiyuan Liu, Sihang Li, Yancheng 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. ArXiv, abs/2310.12798, 2023. URL https://api.semanticscholar. org/CorpusID:264306303

    arXiv 2023

Show all 73 references
  1. [9]

    Instructmol: Multi-modal integration for building a versatile and reliable molecular assistant in drug discovery

    He Cao, Zijing Liu, Xingyu Lu, Yuan Yao, and Yu Li. Instructmol: Multi-modal integration for building a versatile and reliable molecular assistant in drug discovery. ArXiv, abs/2311.16208,

    arXiv

  2. [10]

    Git-mol: A multi-modal large language model for molecular science with graph, image, and text

    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. Computers in Biology and Medicine, 171:108073, March 2024. ISSN 0010-4825. doi: 10.1016/j.compbiomed.2024.108073. URL http://dx.do...

    2024

  3. [11]

    Unimot: Unified molecule- text language model with discrete token representation, 2024

    Juzheng Zhang, Yatao Bian, Yongqiang Chen, and Quanming Yao. Unimot: Unified molecule- text language model with discrete token representation, 2024. URL https://arxiv.org/ abs/2408.00863

    2024 arXiv

  4. [12]

    3d-molm: Towards 3d molecule-text interpretation in language models

    Sihang Li, Zhiyuan Liu, Yanchen Luo, Xiang Wang, Xiangnan He, Kenji Kawaguchi, Tat-Seng Chua, and Qi Tian. 3d-molm: Towards 3d molecule-text interpretation in language models. In ICLR, 2024

    2024

  5. [13]

    Smiles, a chemical language and information system

    David Weininger. Smiles, a chemical language and information system. 1. introduction to methodology and encoding rules. J. Chem. Inf. Comput. Sci., 28(1):31–36, feb 1988. ISSN 0095-2338. doi: 10.1021/ci00057a005. URL https://doi.org/10.1021/ci00057a005

    1988 doi

  6. [14]

    Self-referencing embedded strings (selfies): A 100string representation

    Mario Krenn, Florian Häse, AkshatKumar Nigam, Pascal Friederich, and Alan Aspuru- Guzik. Self-referencing embedded strings (selfies): A 100string representation. Machine Learning: Science and Technology , 1(4):045024, October 2020. ISSN 2632-2153. doi: 10.1088/2632-2153/aba947...

    2020 doi

  7. [15]

    Multi-modal molecule structure-text model for text-based retrieval and editing, 2024

    Shengchao Liu, Weili Nie, Chengpeng Wang, Jiarui Lu, Zhuoran Qiao, Ling Liu, Jian Tang, Chaowei Xiao, and Anima Anandkumar. Multi-modal molecule structure-text model for text-based retrieval and editing, 2024. URL https://arxiv.org/abs/2212.10789

    2024 arXiv

  8. [16]

    Molecular contrastive learning of representations via graph neural networks

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

    2022

  9. [17]

    A molecular multimodal foundation model associating molecule graphs with natural language, 2022

    Bing Su, Dazhao Du, Zhao Yang, Yujie Zhou, Jiangmeng Li, Anyi Rao, Hao Sun, Zhiwu Lu, and Ji-Rong Wen. A molecular multimodal foundation model associating molecule graphs with natural language, 2022. URL https://arxiv.org/abs/2209.05481

    2022 arXiv

  10. [18]

    Drugchat: Towards enabling chatgpt- like capabilities on drug molecule graphs, 2023

    Youwei Liang, Ruiyi Zhang, Li Zhang, and Pengtao Xie. Drugchat: Towards enabling chatgpt- like capabilities on drug molecule graphs, 2023. URL https://arxiv.org/abs/2309. 03907

    2023

  11. [19]

    3d-molt5: Towards unified 3d molecule-text modeling with 3d molecular tokenization, 2024

    Qizhi Pei, Lijun Wu, Kaiyuan Gao, Jinhua Zhu, and Rui Yan. 3d-molt5: Towards unified 3d molecule-text modeling with 3d molecular tokenization, 2024. URL https://arxiv.org/ abs/2406.05797

    2024 arXiv

  12. [20]

    Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models, 2023

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models, 2023. URL https: //arxiv.org/abs/2301.12597. 11

    2023 arXiv

  13. [21]

    Strategies for pre-training graph neural networks

    Weihua Hu, Bowen Liu, Joseph Gomes, Marinka Zitnik, Percy Liang, Vijay Pande, and Jure Leskovec. Strategies for pre-training graph neural networks. In International Conference on Learning Representations, 2020

    2020

  14. [22]

    Pure transformers are powerful graph learners

    Jinwoo Kim, Dat Nguyen, Seonwoo Min, Sungjun Cho, Moontae Lee, Honglak Lee, and Se- unghoon Hong. Pure transformers are powerful graph learners. Advances in Neural Information Processing Systems, 35:14582–14595, 2022

    2022

  15. [23]

    Chemdfm: A large language foundation model for chemistry

    Zihan Zhao, Da Ma, Lu Chen, Liangtai Sun, Zihao Li, Yi Xia, Bo Chen, Hongshen Xu, Zichen Zhu, Su Zhu, et al. Chemdfm: A large language foundation model for chemistry. arXiv preprint arXiv:2401.14818, 2024

    2024 arXiv

  16. [24]

    Deeper insights into graph convolutional networks for semi-supervised learning

    Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for semi-supervised learning. In Proceedings of the AAAI conference on artificial intelligence, volume 32, 2018

    2018

  17. [25]

    Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas ...

    2023 arXiv

  18. [26]

    Visual instruction tun- ing

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tun- ing. ArXiv, abs/2304.08485, 2023. URL https://api.semanticscholar.org/CorpusID: 258179774

    2023 arXiv

  19. [27]

    Huang, Nan Xu, Sheng Zhang, Hoifung Poon, and Muhao Chen

    Fei Wang, Wenxuan Zhou, James Y . Huang, Nan Xu, Sheng Zhang, Hoifung Poon, and Muhao Chen. mdpo: Conditional preference optimization for multimodal large language mod- els. ArXiv, abs/2406.11839, 2024. URL https://api.semanticscholar.org/CorpusID: 270560448

    2024 arXiv

  20. [28]

    Simpo: Simple preference optimization with a reference-free reward

    Yu Meng, Mengzhou Xia, and Danqi Chen. Simpo: Simple preference optimization with a reference-free reward. ArXiv, abs/2405.14734, 2024. URL https://api.semanticscholar. org/CorpusID:269983560

    2024 arXiv

  21. [29]

    Edwards, T

    Carl N. Edwards, T. Lai, Kevin Ros, Garrett Honke, and Heng Ji. Translation between molecules and natural language. ArXiv, abs/2204.11817, 2022. URL https://api.semanticscholar. org/CorpusID:248376906

    2022 arXiv

  22. [30]

    Nomenclature of organic chemistry: IUPAC recommenda- tions and preferred names 2013

    Henri A Favre and Warren H Powell. Nomenclature of organic chemistry: IUPAC recommenda- tions and preferred names 2013. Royal Society of Chemistry, 2013

    2013

  23. [31]

    Hartshorn, Elvis Saravia, Andrew Poulton, Viktor Kerkez, and Robert Stojnic

    Ross Taylor, Marcin Kardas, Guillem Cucurull, Thomas Scialom, Anthony S. Hartshorn, Elvis Saravia, Andrew Poulton, Viktor Kerkez, and Robert Stojnic. Galactica: A large language model for science. ArXiv, abs/2211.09085, 2022. URL https://api.semanticscholar. org/CorpusID:253553203

    2022 arXiv

  24. [32]

    Aqsoldb, a curated reference set of aqueous solubility and 2d descriptors for a diverse set of compounds

    Murat Cihan Sorkun, Abhishek Khetan, and Süleyman Er. Aqsoldb, a curated reference set of aqueous solubility and 2d descriptors for a diverse set of compounds. Scientific Data, 6, 2019. URL https://api.semanticscholar.org/CorpusID:199491456

    2019

  25. [33]

    Wigh, Joe Arrowsmith, Alexander Pomberger, Kobi C

    Daniel S. Wigh, Joe Arrowsmith, Alexander Pomberger, Kobi C. Felton, and Alexei A. Lapkin. Orderly: Data sets and benchmarks for chemical reaction data. Journal of Chemical Infor- mation and Modeling, 64:3790 – 3798, 2024. URL https://api.semanticscholar.org/ CorpusID:269325115

    2024

  26. [34]

    A novel measure for evaluating classifiers

    Jinmao Wei, Xiao-Jie Yuan, Qinghua Hu, and Shuqin Wang. A novel measure for evaluating classifiers. Expert Syst. Appl., 37:3799–3809, 2010. URL https://api.semanticscholar. org/CorpusID:9240275

    2010

  27. [35]

    Bleu: a method for automatic evaluation of machine translation

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311–318, 2002

    2002

  28. [36]

    Rdkit documentation

    Greg Landrum. Rdkit documentation. Release, 1(1-79):4, 2013. 12

    2013

  29. [37]

    Reoptimization of mdl keys for use in drug discovery

    Joseph L Durant, Burton A Leland, Douglas R Henry, and James G Nourse. Reoptimization of mdl keys for use in drug discovery. Journal of chemical information and computer sciences, 42 (6):1273–1280, 2002

    2002

  30. [38]

    The generation of a unique machine description for chemical structures-a technique developed at chemical abstracts service

    Harry L Morgan. The generation of a unique machine description for chemical structures-a technique developed at chemical abstracts service. Journal of chemical documentation, 5(2): 107–113, 1965

    1965

  31. [39]

    Rouge: A package for automatic evaluation of summaries

    Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. InText summarization branches out, pages 74–81, 2004

    2004

  32. [40]

    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. In Proceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, pages 65–72, 2005

    2005

  33. [41]

    Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S

    Zhenqin Wu, Bharath Ramsundar, Evan N. Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S. Pappu, Karl Leswing, and Vijay S. Pande. Moleculenet: a benchmark for molecular machine learning. Chemical Science, 9:513 – 530, 2017. URL https://api.semanticscholar. org/CorpusID:217680306

    2017

  34. [42]

    Graph contrastive learning with augmentations

    Yuning You, Tianlong Chen, Yongduo Sui, Ting Chen, Zhangyang Wang, and Yang Shen. Graph contrastive learning with augmentations. Advances in neural information processing systems, 33:5812–5823, 2020

    2020

  35. [43]

    Exploring the limits of transfer learning with a unified text-to-text transformer

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research, 21(140):1–67, 2020

    2020

  36. [44]

    Molxpt: Wrapping molecules with text for generative pre-training

    Zequn Liu, Wei Zhang, Yingce Xia, Lijun Wu, Shufang Xie, Tao Qin, Ming Zhang, and Tie-Yan Liu. Molxpt: Wrapping molecules with text for generative pre-training. In ACL, 2023

    2023

  37. [45]

    Language models are unsupervised multitask learners

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019

    2019

  38. [46]

    Gerstein

    Xiangru Tang, Andrew Tran, Jeffrey Tan, and Mark B. Gerstein. Mollm: a unified language model for integrating biomedical text with 2d and 3d molecular representations. Bioinformatics, 40:i357 – i368, 2024. URL https://api.semanticscholar.org/CorpusID:265455405

    2024

  39. [47]

    Direct preference optimization: Your language model is secretly a reward model

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors, Advances in N...

    2023

  40. [48]

    Detecting and mitigating hallucination in large vision language models via fine-grained ai feedback

    Wenyi Xiao, Ziwei Huang, Leilei Gan, Wanggui He, Haoyuan Li, Zhelun Yu, Hao Jiang, Fei Wu, and Linchao Zhu. Detecting and mitigating hallucination in large vision language models via fine-grained ai feedback. ArXiv, abs/2404.14233, 2024. URL https://api. semanticscholar.org/Co...

    2024 arXiv

  41. [49]

    Aligning modalities in vision large language models via preference fine-tuning

    Yiyang Zhou, Chenhang Cui, Rafael Rafailov, Chelsea Finn, and Huaxiu Yao. Aligning modalities in vision large language models via preference fine-tuning. ArXiv, abs/2402.11411,

    arXiv

  42. [50]

    Strengthening multimodal large language model with bootstrapped preference optimiza- tion

    Renjie Pi, Tianyang Han, Wei Xiong, Jipeng Zhang, Runtao Liu, Rui Pan, and Tong Zhang. Strengthening multimodal large language model with bootstrapped preference optimiza- tion. ArXiv, abs/2403.08730, 2024. URL https://api.semanticscholar.org/CorpusID: 268379605

    2024 arXiv

  43. [51]

    Enhancing large vision language models with self-training on image comprehen- sion

    Yihe Deng, Pan Lu, Fan Yin, Ziniu Hu, Sheng Shen, James Zou, Kai-Wei Chang, and Wei Wang. Enhancing large vision language models with self-training on image comprehen- sion. ArXiv, abs/2405.19716, 2024. URL https://api.semanticscholar.org/CorpusID: 270123045. 13

    2024 arXiv

  44. [52]

    Scibert: A pretrained language model for scientific text

    Iz Beltagy, Kyle Lo, and Arman Cohan. Scibert: A pretrained language model for scientific text. In Conference on Empirical Methods in Natural Language Processing , 2019. URL https://api.semanticscholar.org/CorpusID:202558505

    2019

  45. [53]

    Text2mol: Cross-modal molecule retrieval with natural language queries

    Carl Edwards, ChengXiang Zhai, and Heng Ji. Text2mol: Cross-modal molecule retrieval with natural language queries. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 595–607, 2021

    2021

  46. [54]

    Shoemaker, Paul A

    Sunghwan Kim, Jie Chen, Tiejun Cheng, Asta Gindulyte, Jia He, Siqian He, Qingliang Li, Benjamin A. Shoemaker, Paul A. Thiessen, Bo Yu, Leonid Y . Zaslavsky, Jian Zhang, and Evan E. Bolton. Pubchem 2023 update. Nucleic acids research, 2022. URL https://api. semanticscholar.org/...

    2023

  47. [57]

    datasets. Model BLEU-2 ( ↑) BLEU-4 (↑) ROUGE-1 (↑) ROGUE-2 (↑) ROUGE-L (↑) METEOR (↑) ChEBI-20 Specialist Models GIT-Mol 0.352 0.263 0.575 0.485 0.560 0.533 InstructMol 0.475 0.371 0.566 0.394 0.502 0.509 MolT5 0.594 0.508 0.654 0.510 0.594 0.614 MolCA∗ 0.623 0.540 0.693 0.553...

  48. [58]

    Guidelines: • The answer NA means that the abstract and introduction do not include the claims made in the paper

    Claims Question: Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? Answer: [Yes] Justification: The claims have been validated through extensive evaluation on various molecular tasks in Section 3 and Appendix D.3, ...

  49. [59]

    Limitations

    Limitations Question: Does the paper discuss the limitations of the work performed by the authors? Answer: [Yes] Justification: We discuss limitations in Section 5 and appendix A. Guidelines: • The answer NA means that the paper has no limitation while the answer No means that...

  50. [60]

    24 Guidelines: • The answer NA means that the paper does not include theoretical results

    Theory assumptions and proofs Question: For each theoretical result, does the paper provide the full set of assumptions and a complete (and correct) proof? Answer: [NA] Justification: The paper does not include theoretical results. 24 Guidelines: • The answer NA means that the...

  51. [61]

    Appendix B includes further training details

    Experimental result reproducibility Question: Does the paper fully disclose all the information needed to reproduce the main ex- perimental results of the paper to the extent that it affects the main claims and/or conclusions of the paper (regardless of whether the code and da...

  52. [62]

    In addition, the code, trained model, and data will be released publicly

    Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instruc- tions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] 25 Justification: Appendix B provides link...

  53. [63]

    Experimental setting/details Question: Does the paper specify all the training and test details (e.g., data splits, hyper- parameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: All experimental settings are...

  54. [64]

    We note that the prior work we discussed in Section 4 similarly does not report statistical significance

    Experiment statistical significance Question: Does the paper report error bars suitably and correctly defined or other appropriate information about the statistical significance of the experiments? Answer: [No] Justification: For generalist models, reporting results with such ...

  55. [65]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiments compute resources Question: For each experiment, does the paper provide sufficient information on the com- puter resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Yes] Justification: Appendix D.1 provides s...

  56. [66]

    Guidelines: • The answer NA means that the authors have not reviewed the NeurIPS Code of Ethics

    Code of ethics Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? Answer: [Yes] Justification: The research presented in this paper fully conforms to the NeurIPS Code of Ethi...

  57. [67]

    Guidelines: • The answer NA means that there is no societal impact of the work performed

    Broader impacts Question: Does the paper discuss both potential positive societal impacts and negative societal impacts of the work performed? Answer: [Yes] Justification: We discuss broader impacts in Appendix E. Guidelines: • The answer NA means that there is no societal imp...

  58. [68]

    Guidelines: • The answer NA means that the paper poses no such risks

    Safeguards Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g., pretrained language models, image generators, or scraped datasets)? Answer: [NA] Justification: Our model, Mol-...

  59. [69]

    Guidelines: • The answer NA means that the paper does not use existing assets

    Licenses for existing assets Question: Are the creators or original owners of assets (e.g., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? Answer: [Yes] Justification: The backbone mod...

  60. [70]

    Guidelines: • The answer NA means that the paper does not release new assets

    New assets Question: Are new assets introduced in the paper well documented and is the documentation provided alongside the assets? Answer: [Yes] Justification: We release our model, code, and data along with comprehensive documentation, which is available in the code reposito...

  61. [71]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Crowdsourcing and research with human subjects Question: For crowdsourcing experiments and research with human subjects, does the paper include the full text of instructions given to participants and screenshots, if applicable, as well as details about compensation (if any)? A...

  62. [72]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals...

  63. [73]

    29 Answer: [Yes] Justification: Large language models (LLMs) are an integral component of the proposed method, as described in Section 2

    Declaration of LLM usage Question: Does the paper describe the usage of LLMs if it is an important, original, or non-standard component of the core methods in this research? Note that if the LLM is used only for writing, editing, or formatting purposes and does not impact the ...

    2025

  64. [2023]

    URL https://api.semanticscholar.org/CorpusID:265466509

  65. [2024]

    URL https://api.semanticscholar.org/CorpusID:267750239

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.