Pith. sign in

REVIEW 4 major objections 7 minor 41 references

MTPNet: Multi-Grained Target Perception for Unified Activity Cliff Prediction

T0 review · 4 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read MTPNet conditions molecular representations on the receptor protein—at whole-target and binding-pocket granularity—and reports average activity-cliff error reduced by 18.95% across 30 benchmark datasets.

desk verdict Receptor-conditioned activity cliff prediction is a useful idea, but the reported RMSE gains are confounded by joint training and target identity until matched baselines are added. read the letter →

arxiv 2506.05427 v1 pith:G4DTOP4E submitted 2025-06-05 cs.LG cs.AIcs.CE

Zishan Shu , Yufan Deng , Hongyu Zhang , Zhiwei Nie , Jie Chen This is my paper
classification cs.LGcs.AIcs.CE
keywords activitycliffpredictiondrugdiscoverygraphneuralnetworksproteinlanguagemodelcross-attentionadaptivelayernormalizationreceptor-ligandinteractionsMoleculeACE
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

Activity cliffs are the drug designer's trap: two molecules that differ by a single substituent can differ wildly in potency, and standard graph-neural-network models that look only at the molecule routinely miss the jump. This paper claims that the missing information is the protein the molecule binds. It proposes MTPNet, a unified network that conditions the molecular representation on the receptor at two granularities—whole-target semantics via adaptive layer normalization, and binding-pocket details via cross-attention—and reports that on the 30 MoleculeACE datasets this cuts average RMSE by 18.95% on top of mainstream GNN backbones, with the same model reaching AUC 0.924 on CYP3A4 cliff classification. If the claim holds, a single target-aware model could replace the per-target models that dominate activity cliff prediction today, and the attention maps would double as a chemical rationale for each prediction.

What carries the argument

The load-bearing object is the Multi-Grained Target Perception (MTP) module, a stacked attention block that iteratively refines the ligand feature matrix $F_{\mathrm{mol}}$ under two alternating protein conditions. Macro-level Target Semantic (MTS) guidance compresses the target features by average pooling, sends the compressed vector through a linear Weight Regressor to produce dynamic weights $(\gamma_i, \beta_i)$, and uses them as Adaptive Layer Normalization (AdaLN) parameters before a self-attention layer, aligning the molecule with the whole receptor's semantic context. Micro-level Pocket Semantic (MPS) guidance extracts a binding-pocket feature matrix $F_{\mathrm{pocket}}$ from the target using a Pocket Prior Extractor such as Cavity Plus, then runs scaled dot-product cross-attention with the molecule features as the query and the pocket features as key and value, capturing local receptor-ligand contacts. Alternating the two mechanisms across $L$ stacked layers lets the network internalize both global interaction patterns and fine-grained structural sensitivity, which the paper argues is exactly what an activity cliff—a small molecular change producing a large activity change—requires.

What would settle it

Rerun the plug-and-play comparison that produces the headline 19% RMSE gain, but replace the pocket prior with the same number of residues sampled uniformly at random from each target's sequence; if RMSE stays near the reported MTPNet values instead of drifting back toward the molecule-only baselines, pocket semantics are not the operative ingredient. A complementary check is to verify that all 30 MoleculeACE targets actually have a 3D structure or cavity annotation available, since the micro-level branch has no defined input for a target without one.

Watch

Extended reading notes

Core claim

The paper's central claim is that receptor proteins can be used as conditional guidance, not just as side information, to improve activity cliff prediction; it states this is the first time receptor proteins are employed as guiding information to capture critical interaction details. MTPNet embeds each ligand with Mole-BERT and each target with ESM2, then runs a Multi-Grained Target Perception (MTP) module that alternates Macro-level Target Semantic guidance (average-pooled target features turned by a weight regressor into adaptive layer-normalization parameters gamma and beta) and Micro-level Pocket Semantic guidance (binding-pocket features from a Pocket Prior Extractor such as Cavity Plus, fused with the molecule by cross-attention). Through stacked layers the ligand representation is progressively aligned to the receptor, and the same network serves single-target and multi-target datasets. The paper reports that MTPNet beats all baselines on all 30 MoleculeACE datasets with an average RMSE improvement of 18.95% over mainstream GNN architectures, that the improvement is not explained by parameter scaling (scaling GCN's parameters from 1.11M to 3.17M only reduces RMSE from 0.950 to 0.915, while adding MTP reaches 0.744), and that the learned attention concentrates on functional groups and double or triple bonds, aligning with known chemistry.

Load-bearing premise

The load-bearing premise is that a trustworthy binding pocket can be found for every target protein and turned into features compatible with the protein encoder; the paper never says whether 3D structures exist for all 30 datasets, how pocket residues are chosen, or how they are aligned with the sequence features—if pocket extraction is noisy or unavailable, the cross-attention is conditioning on missing or approximate inputs, and the reported gains could come from extra parameters or from the whole-target guidance alone.

Editorial extensions

If this is right

  • One unified network, MTPNet, predicts activity cliffs across all 30 MoleculeACE targets instead of requiring a separate model per binding target, eliminating the per-target redundancy of current approaches.
  • Adding the MTP module to six mainstream GNN backbones (GCN, GAT, GIN, GraphTrans, MolCLR, Mole-BERT) improves PCC by 11.6%, R-squared by 17.8%, and RMSE by 19.0% on average, and the scale-up control shows that simply enlarging the baselines recovers only a small fraction of that gain.
  • The same receptor-conditioned model transfers to classification, reaching AUC 0.924 on the CYP3A4 cliff dataset, above Mole-BERT (0.902) and MolCLR (0.896).
  • MTPNet's attention maps concentrate on functional groups (amino, carbonyl, sulfonyl, carboxyl, halogen) and on double and triple bonds, so the predictions come with a chemical rationale that matches known binding chemistry.

Reading between the lines

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

  • A decisive control the paper does not run would be to feed the MTP module randomly chosen pocket residues of the same count; if the RMSE gain over molecule-only baselines survives, pocket semantics are contributing little and the gain should be credited to the macro guidance or to added capacity.
  • If receptor conditioning is genuinely the driver, the same conditioning idea should transfer to other target-dependent endpoints such as matched molecular pairs, solubility, or toxicity panels, and the cheapest probe would be swapping the ESM2 and Mole-BERT embeddings for plain fingerprints to see how much of the 18.95% depends on pretrained representations.
  • The paper's 'first time' framing implies molecule-only pretraining has plateaued on MoleculeACE; if that is so, the reported numbers set an upper bound on what receptor conditioning alone can buy on this benchmark, and combining it with larger protein language models is the natural next step.
  • Whether pocket annotations exist for all 30 targets is left open; a testable prediction is that MTPNet's performance degrades smoothly as pocket-residue noise is injected, which would locate the mechanism in pocket fidelity rather than in the attention machinery.
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

4 major / 7 minor

Summary. The paper proposes MTPNet, a conditional framework that augments molecular GNNs with receptor protein information at two levels: Macro-level Target Semantic (MTS) guidance, which uses adaptive layer normalization conditioned on pooled target features, and Micro-level Pocket Semantic (MPS) guidance, which uses cross-attention between molecule and pocket features. The authors report experiments on 30 MoleculeACE datasets, plug-and-play gains over six GNN backbones, an ablation study, a CYP3A4 classification result, and an attention visualization. The central claim is that receptor/pocket guidance yields large RMSE improvements and that MTPNet is the first unified, target-aware activity-cliff predictor.

Significance. If the claimed gains are reproducible and can be attributed to protein/pocket conditioning rather than to target identity, joint training, or added parameters, the paper would make a useful contribution: it offers a pluggable module for existing GNNs and a unified framework for activity-cliff prediction. The code release is a strength, and the idea of combining global target semantics with local pocket semantics is reasonable. However, the current evidence does not yet establish that protein semantics, rather than target identity or training protocol, explains the reported improvements, so the significance of the central claim remains conditional on additional experiments.

major comments (4)
  1. [Section 4.2, Table 1, Table 3] The paper never states whether the baseline models (GCN, GAT, GIN, GraphTrans, MolCLR, Mole-BERT) are trained per target, as in the standard MoleculeACE benchmark, while MTPNet is trained jointly across the 30 datasets as a 'unified' model. Within each MoleculeACE dataset the protein is fixed, so F_target is a constant vector and can act as a dataset/target embedding. A jointly trained molecule-only backbone (Table 3, 'w/o MTS & MPS', RMSE 0.917) is not shown to be matched to this protocol, and the scale-up control in Table 1 varies parameters but not target identity or training protocol. To support the central claim that receptor/pocket information drives the 18.95% improvement, the authors should report per-target results and matched baselines, including a per-target-trained MTPNet, a jointly trained molecule-only backbone, and a control with a one-hot target-ID embedding or per-target bias in place of F_target.
  2. [Table 1 and Table 3] Only aggregate RMSE, PCC, and R² values are reported, with no per-dataset breakdown, error bars, or significance tests. The three headline improvements are mutually inconsistent: the abstract and Section 4.2 state an average RMSE improvement of 18.95% over mainstream GNN architectures, the introduction states a 7.2% improvement over SOTA, and the plug-and-play results in Table 1 are summarized as a 19.0% RMSE decrease. Additionally, Table 3's 'w/o MTS & MPS' baseline (RMSE 0.917) does not match the Mole-BERT baseline in Table 1 (RMSE 0.947), so the reader cannot tell whether the two tables use the same training protocol. Per-dataset tables, confidence intervals, and a precise definition of each reported improvement are needed before the claim of significant improvement can be evaluated.
  3. [Section 3.2c] The pocket feature extraction is underspecified. The text states only that F_pocket ∈ R^{p×d} is extracted from F_target using a Pocket Prior Extractor (e.g., Cavity Plus), without saying whether 3D structures are available for all 30 MoleculeACE targets, how pocket residues are selected, how p is set, or how these residue embeddings are projected into the same space as ESM2 sequence features. If pocket extraction is noisy or unavailable for some targets, the MPS cross-attention operates on missing or approximate inputs, and the reported gains could come from added parameters or from MTS alone. The authors should specify the exact pocket extraction pipeline and include an ablation that removes MPS while keeping the same parameter count.
  4. [Equations (6), (12), and (13)] There are internal inconsistencies in the MPS formulation. Eq. (6) calls Φ_MPS with F_target, while Eq. (13) defines Φ_MPS(F_mol, F_target) = CrossAttention(F_mol, F_pocket); moreover, the text says K_pocket and V_pocket are generated by concatenating F_mol and F_pocket, but Eq. (12) defines K_pocket = W_k F_pocket and V_pocket = W_v F_pocket with no concatenation. These mismatches make the exact micro-level mechanism ambiguous and need to be corrected.
minor comments (7)
  1. [Section 1] The word 'summaried' in the contributions list should be 'summarized.'
  2. [Section 3.2b] The Feature Abstractor is said to 'reduce the feature dimensions' via average pooling; since F_target is n×d, average pooling over the n dimension gives a 1×d vector, so the description should clarify whether the reduction is in sequence length or in d.
  3. [Section 4.2] The text refers to results presented in 'Appendix,' but no appendix is included in the manuscript; either add the per-dataset tables or remove the reference.
  4. [Section 3.2c] The cross-attention text says F_mol and F_pocket are concatenated to generate K_pocket and V_pocket, but the equation shows only F_pocket; the discrepancy should be resolved.
  5. [Figure 1 caption] The caption says RMSE values are shown in reciprocal form, but the axis labels and legend are not explained; please state what the plotted quantity is and how the 7.2% improvement is computed.
  6. [Section 4.4] The interpretation analysis is qualitative and does not provide any quantitative comparison (e.g., attention-overlap scores or performance on molecules with and without the highlighted groups); please add a quantitative measure or soften the claim.
  7. [Section 4.1 and Table 2] The paper does not report hyperparameters, training epochs, data splits, or number of random seeds for the CYP3A4 classification experiment; please add these details to allow reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: MTPNet's predictive gains are measured on held-out data; the MTS/MPS conditioning is computed from external protein/pocket features, not derived from the activity labels.

full rationale

The paper's central claims are empirical supervised predictions evaluated on held-out MoleculeACE datasets and an external CYP3A4 classification benchmark. MTPNet is a conditional architecture: MTS guidance computes conditional weights via AvgPool and linear projections of target features (Eqs. 8-10), and MPS guidance extracts pocket features from the target representation and applies cross-attention (Eqs. 12-13). These are input features, not fitted targets. No parameter is fit to the test labels, and no prediction is restated from the training data by construction. The reported RMSE/PCC/R2 improvements come from comparisons between models under the same evaluation protocol; the ablation rows (w/o MTS, w/o MPS) are held-out comparisons, not algebraic identities. Self-citations to the authors' prior work (Nie et al. 2024a,b; Wu et al. 2025) appear only in general related-work statements about GNNs, protein language models, and peptide binding; none is invoked as a uniqueness theorem or as the basis for the MTP module design. The 'first time to employ receptor proteins as guiding information' claim is a novelty assertion, not a derived result. Potential concerns raised by a skeptical reading, such as whether baselines were trained with the same joint multi-target protocol or whether pocket extraction is noisy, are experimental confounds or ablations, not circularity. The derivation chain—from molecular and protein inputs through a learned conditional module to a regression/classification output—is self-contained against external benchmarks and does not reduce to its inputs by definition. Therefore no significant circularity is present.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on pretrained protein and molecule encoders, on the assumption that pocket features can be extracted reliably, and on fair benchmark comparison. The model's own weights are fitted to data, which is standard, but the paper does not report hyperparameters or the pocket extraction details, so the contribution as stated cannot be fully audited.

free parameters (3)
  • MTP module learnable weights (W_q, W_k, W_v, AdaLN, Weight Regressor) = not reported
    The cross-attention and adaptive normalization parameters are fit to training data; their exact values are not provided, so the reported improvements cannot be audited.
  • Number of MTP layers L = not reported
    The depth of the stacked MTS/MPS refinement is a hand-chosen hyperparameter; the paper does not state its value.
  • Embedding dimension d and pocket residue count p = not reported
    These determine the size of F_target and F_pocket; without them, the architecture is underspecified.
assumptions (5)
  • domain assumption ESM2 and Mole-BERT pretrained representations encode sufficient chemical and biological signal for activity cliff prediction
    The method uses these pretrained features as input; if they are uninformative, the MTP module cannot add value. Invoked in Section 3.2.
  • domain assumption A reliable binding pocket can be obtained for each target, and pocket residues can be mapped to ESM2 sequence embeddings
    The MPS guidance depends on F_pocket extracted 'e.g., Cavity Plus'; the paper gives no details of structure availability or residue mapping (Section 3.2c).
  • domain assumption MoleculeACE benchmark splits are respected and baseline implementations are comparable
    The empirical claim rests on fair comparison; the paper does not state whether standard splits were used or whether baselines were retrained with the same protocol (Section 4.1).
  • domain assumption Activity cliffs are driven by receptor-ligand interactions and can be captured by conditioning molecular representations on protein semantics
    This is the conceptual premise of the entire framework, stated in Section 1 and Figure 2.
  • ad hoc to paper Linear transformations and cross-attention can approximate receptor-ligand interaction patterns
    The specific MTS/MPS functional forms are chosen for this task; no universal approximation or theoretical guarantee is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MTPNet: Multi-Grained Target Perception for Unified Activity Cliff Prediction." pith.science (2026). https://pith.science/paper/G4DTOP4E

@misc{pith2026250605427,
  author       = {Pith},
  title        = {Pith review of: MTPNet: Multi-Grained Target Perception for Unified Activity Cliff Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G4DTOP4E}},
  note         = {Machine review of arXiv:2506.05427}
}
read the original abstract

Activity cliff prediction is a critical task in drug discovery and material design. Existing computational methods are limited to handling single binding targets, which restricts the applicability of these prediction models. In this paper, we present the Multi-Grained Target Perception network (MTPNet) to incorporate the prior knowledge of interactions between the molecules and their target proteins. Specifically, MTPNet is a unified framework for activity cliff prediction, which consists of two components: Macro-level Target Semantic (MTS) guidance and Micro-level Pocket Semantic (MPS) guidance. By this way, MTPNet dynamically optimizes molecular representations through multi-grained protein semantic conditions. To our knowledge, it is the first time to employ the receptor proteins as guiding information to effectively capture critical interaction details. Extensive experiments on 30 representative activity cliff datasets demonstrate that MTPNet significantly outperforms previous approaches, achieving an average RMSE improvement of 18.95% on top of several mainstream GNN architectures. Overall, MTPNet internalizes interaction patterns through conditional deep learning to achieve unified predictions of activity cliffs, helping to accelerate compound optimization and design. Codes are available at: https://github.com/ZishanShu/MTPNet.

Figures

Figures reproduced from arXiv: 2506.05427 by the authors.

Figure 1
Figure 1. The overall RMSE performance comparison of various [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Motivation for incorporating receptor protein informa [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The overview of MTPNet. (a) The complete workflow of MTPNet, beginning with the embedding of molecules and target proteins, [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Visualization of Atom Attention Scores for Molecules in CHEMBL v29. Figure 4 compares the atom attention distributions [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 40 canonical work pages

  1. [1]

    Proteinbert: a uni- versal deep-learning model of protein sequence and func- tion.Bioinformatics, 38(8):2102–2110,

    [Brandeset al., 2022 ] Nadav Brandes, Dan Ofer, Yam Peleg, Nadav Rappoport, and Michal Linial. Proteinbert: a uni- versal deep-learning model of protein sequence and func- tion.Bioinformatics, 38(8):2102–2110,

    work page 2022

  2. [4]

    [Consonni and Todeschini, 2010] Viviana Consonni and Roberto Todeschini.Molecular Descriptors, pages 29–102. Jan

    work page 2010

  3. [6]

    Drucker, C

    [Druckeret al., 1996 ] H. Drucker, C. J. Burges, L. Kaufman, A. Smola, and V . Vapnik. Support vector regression ma- chines. InAdvances in Neural Information Processing Sys- tems, volume 9,

    work page 1996

  4. [13]

    Evolutionary-scale pre- diction of atomic-level protein structure with a language model.Science, 379(6637):1123–1130,

    [Linet al., 2023 ] Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Nikita Smetanin, Robert Verkuil, Ori Kabeli, Yaniv Shmueli, Allan dos San- tos Costa, Maryam Fazel-Zarandi, Tom Sercu, Salvatore Candido, and Alexander Rives. Evolutionary-scale pre- diction of atomic-level protein structure with a language model.Science, 379(6637)...

    work page 2023

  5. [14]

    Prollama: A protein large language model for multi-task protein language processing.IEEE Transactions on Artificial Intelligence,

    [Lvet al., 2025 ] Liuzhenghao Lv, Zongying Lin, Hao Li, Yuyang Liu, Jiaxi Cui, Calvin Yu-Chian Chen, Li Yuan, and Yonghong Tian. Prollama: A protein large language model for multi-task protein language processing.IEEE Transactions on Artificial Intelligence,

    work page 2025

  6. [15]

    Towards geo- metric normalization techniques in se (3) equivariant graph neural networks for physical dynamics simulations

    [Menget al., 2024 ] Ziqiao Meng, Liang Zeng, Zixing Song, Tingyang Xu, Peilin Zhao, and Irwin King. Towards geo- metric normalization techniques in se (3) equivariant graph neural networks for physical dynamics simulations. In Proceedings of the Thirty-Third International Joint Con- ference on Artificial Intelligence, pages 5981–5989,

    work page 2024

  7. [16]

    Mordred: a molecular descriptor calculator.Journal of Cheminformat- ics, Dec

    [Moriwakiet al., 2018 ] Hirotomo Moriwaki, Yu-Shi Tian, Norihito Kawashita, and Tatsuya Takagi. Mordred: a molecular descriptor calculator.Journal of Cheminformat- ics, Dec

    work page 2018

  8. [17]

    Hunting for peptide binders of spe- cific targets with data-centric generative language models

    [Nieet al., 2024a ] Zhiwei Nie, Daixi Li, Jie Chen, Fan Xu, Yutian Liu, Jie Fu, Xudong Liu, Zhennan Wang, Yiming Ma, Kai Wang, et al. Hunting for peptide binders of spe- cific targets with data-centric generative language models. bioRxiv, pages 2023–12,

    work page 2023

Show all 41 references
  1. [18]

    Acgcn: Graph convolutional networks for activity cliff prediction be- tween matched molecular pairs.Journal of Chemical In- formation and Modeling, 62(10):2341–2351,

    [Parket al., 2022 ] Junhui Park, Gaeun Sung, SeungHyun Lee, SeungHo Kang, and ChunKyun Park. Acgcn: Graph convolutional networks for activity cliff prediction be- tween matched molecular pairs.Journal of Chemical In- formation and Modeling, 62(10):2341–2351,

    2022

  2. [19]

    Quantitative evaluation of explainable graph neural networks for molecular property prediction

    [Raoet al., 2022 ] Jiahua Rao, Shuangjia Zheng, Yutong Lu, and Yuedong Yang. Quantitative evaluation of explainable graph neural networks for molecular property prediction. Patterns, 3(12):100628,

    2022

  3. [20]

    Self-supervised graph transformer on large-scale molecular data.arXiv: Biomolecules,arXiv: Biomolecules, Jun

    [Ronget al., 2020 ] Yu Rong, Yatao Bian, Tingyang Xu, Weiyang Xie, Ying Wei, Wenbing Huang, and Jun- zhou Huang. Self-supervised graph transformer on large-scale molecular data.arXiv: Biomolecules,arXiv: Biomolecules, Jun

    2020

  4. [22]

    Saprot: Protein language modeling with structure-aware vocabulary, Oct

    [Suet al., 2023 ] Jin Su, Chenchen Han, Yuyang Zhou, Jun- jie Shan, Xibin Zhou, and Fajie Yuan. Saprot: Protein language modeling with structure-aware vocabulary, Oct

    2023

  5. [23]

    [Van Tilborget al., 2022] Derek Van Tilborg, Alisa Alenicheva, and Francesca Grisoni

    bioRxiv 2023.10.01.560349. [Van Tilborget al., 2022] Derek Van Tilborg, Alisa Alenicheva, and Francesca Grisoni. Exposing the limitations of molecular machine learning with activity cliffs.Journal of chemical information and modeling, 62(23):5938–5951,

    2023

  6. [24]

    Springer Science & Business Media,

    [Vapnik, 2013] Vladimir Vapnik.The Nature of Statistical Learning Theory. Springer Science & Business Media,

    2013

  7. [27]

    Bryant, Tiejun Cheng, Jiyao Wang, Asta Gindulyte, Benjamin A

    [Wanget al., 2017 ] Yanli Wang, Stephen H. Bryant, Tiejun Cheng, Jiyao Wang, Asta Gindulyte, Benjamin A. Shoe- maker, Paul A. Thiessen, Siqian He, and Jian Zhang. Pub- chem bioassay: 2017 update.Nucleic Acids Research, page D955–D963, Jan

    2017

  8. [28]

    [Wanget al., 2022 ] Y . Wang, J. Wang, Z. Cao, et al. Molecu- lar contrastive learning of representations via graph neural networks.Nature Machine Intelligence, 4:279–287,

    2022

  9. [29]

    Representing long-range context for graph neu- ral networks with global attention

    [Wuet al., 2021 ] Zhanghao Wu, Paras Jain, Matthew Wright, Azalia Mirhoseini, Joseph E Gonzalez, and Ion Stoica. Representing long-range context for graph neu- ral networks with global attention. In M. Ranzato, A. Beygelzimer, Y . Dauphin, P.S. Liang, and J. Wortman Vaughan, e...

    2021

  10. [30]

    Re- thinking text-based protein understanding: Retrieval or llm?,

    [Wuet al., 2025 ] Juntong Wu, Zijing Liu, He Cao, Hao Li, Bin Feng, Zishan Shu, Ke Yu, Li Yuan, and Yu Li. Re- thinking text-based protein understanding: Retrieval or llm?,

    2025

  11. [31]

    A semi-supervised molecular learning framework for activity cliff estimation

    [Wu, 2024] Fang Wu. A semi-supervised molecular learning framework for activity cliff estimation. InProceedings of the Thirty-Third International Joint Conference on Artifi- cial Intelligence, pages 6080–6088,

    2024

  12. [32]

    [Xiaet al., 2023b ] Jun Xia, Chengshuai Zhao, Bozhen Hu, Zhangyang Gao, Cheng Tan, Yue Liu, Siyuan Li, and Stan Z. Li. Mole-bert: Rethinking pre-training graph neu- ral networks for molecules. InThe Eleventh International Conference on Learning Representations, ICLR 2023, Ki- ...

    2023

  13. [33]

    An image-enhanced molecular graph representation learn- ing framework

    [Xianget al., 2024 ] Hongxin Xiang, Shuting Jin, Jun Xia, Man Zhou, Jianmin Wang, Li Zeng, and Xiangxiang Zeng. An image-enhanced molecular graph representation learn- ing framework. InProceedings of the Thirty-Third Inter- national Joint Conference on Artificial Intelligence,...

    2024

  14. [34]

    Deepprotein: Deep learning library and benchmark for protein sequence learning,

    [Xieet al., 2024 ] Jiaqing Xie, Yue Zhao, and Tianfan Fu. Deepprotein: Deep learning library and benchmark for protein sequence learning,

    2024

  15. [35]

    [Xuet al., 2018 ] Y Xu, S Wang, Q Hu, S Gao, X Ma, W Zhang, Y Shen, F Chen, L Lai, and J Pei. Cavity- plus: a web server for protein cavity detection with phar- macophore modelling, allosteric site identification and co- valent ligand binding ability prediction.Nucleic Acids R...

    2018

  16. [36]

    How powerful are graph neural net- works? InInternational Conference on Learning Rep- resentations,

    [Xuet al., 2019 ] Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural net- works? InInternational Conference on Learning Rep- resentations,

    2019

  17. [38]

    A pre-trained multi-representation fusion network for molecular property prediction.Infor- mation Fusion, 103:102092,

    [Zhanget al., 2024 ] Haohui Zhang, Juntong Wu, Shichao Liu, and Shen Han. A pre-trained multi-representation fusion network for molecular property prediction.Infor- mation Fusion, 103:102092,

    2024

  18. [39]

    Semignn-ppi: Self-ensembling multi-graph neural network for efficient and generalizable protein-protein in- teraction prediction,

    [Zhaoet al., 2023 ] Ziyuan Zhao, Peisheng Qian, Xulei Yang, Zeng Zeng, Cuntai Guan, Wai Leong Tam, and Xi- aoli Li. Semignn-ppi: Self-ensembling multi-graph neural network for efficient and generalizable protein-protein in- teraction prediction,

    2023

  19. [40]

    Cross- view contrastive fusion for enhanced molecular property prediction

    [Zhenget al., 2024 ] Yan Zheng, Song Wu, Junyu Lin, Yazhou Ren, Jing He, Xiaorong Pu, and Lifang He. Cross- view contrastive fusion for enhanced molecular property prediction. InProccedings of the Thirty-Third Interna- tional Joint Conference on Artificial Intelligence,

    2024

  20. [41]

    Molhf: A hierarchical normalizing flow for molecular graph generation, 2023

    [Zhuet al., 2023 ] Yiheng Zhu, Zhenqiu Ouyang, Ben Liao, Jialu Wu, Yixuan Wu, Chang-Yu Hsieh, Tingjun Hou, and Jian Wu. Molhf: A hierarchical normalizing flow for molecular graph generation, 2023

    2023

  21. [1996]

    Mmgnn: A molecular merged graph neural network for explainable solvation free energy prediction

    [Duet al., 2024 ] Wenjie Du, Shuai Zhang, Jun Xia Di Wu, Ziyuan Zhao, Junfeng Fang, and Yang Wang. Mmgnn: A molecular merged graph neural network for explainable solvation free energy prediction. InProceedings of the Thirty-Third International Joint Conference on Artificial In...

    2024

  22. [2009]

    Graph attention networks.arXiv preprint arXiv:1710.10903,

    [Veliˇckovi´cet al., 2017 ] Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. Graph attention networks.arXiv preprint arXiv:1710.10903,

    2017 arXiv

  23. [2010]

    Admetlab: a platform for sys- tematic admet evaluation based on a comprehensively col- lected admet database.Journal of Cheminformatics, 10(1), Dec

    [Donget al., 2018 ] Jie Dong, Ning-Ning Wang, Zhi-Jiang Yao, Lin Zhang, Yan Cheng, Defang Ouyang, Ai-Ping Lu, and Dong-Sheng Cao. Admetlab: a platform for sys- tematic admet evaluation based on a comprehensively col- lected admet database.Journal of Cheminformatics, 10(1), Dec

    2018

  24. [2013]

    Austin, David G

    [Veithet al., 2009 ] Henrike Veith, Noel Southall, Ruili Huang, Tim James, Darren Fayne, Natalia Artemenko, Min Shen, James Inglese, Christopher P. Austin, David G. Lloyd, and et al. Comprehensive characterization of cy- tochrome p450 isozyme selectivity across chemical li- br...

    2009

  25. [2017]

    Glpocket: A multi-scale representation learning approach for protein binding site prediction

    [Liet al., 2023 ] Peiying Li, Yongchang Liu, Shikui Tu, and Lei Xu. Glpocket: A multi-scale representation learning approach for protein binding site prediction. InIJCAI, pages 4821–4828,

    2023

  26. [2018]

    Deepac – conditional transformer-based chem- ical language model for the prediction of activity cliffs formed by bioactive compounds.Digital Discovery, 1(6):898–909,

    [Chenet al., 2022 ] Hengwei Chen, Martin V ogt, and J¨urgen Bajorath. Deepac – conditional transformer-based chem- ical language model for the prediction of activity cliffs formed by bioactive compounds.Digital Discovery, 1(6):898–909,

    2022

  27. [2019]

    Gpmo: Gradient perturbation-based contrastive learning for molecule optimization

    [Yanget al., 2023 ] Xixi Yang, Li Fu, Yafeng Deng, Yuan- sheng Liu, Dongsheng Cao, and Xiangxiang Zeng. Gpmo: Gradient perturbation-based contrastive learning for molecule optimization. InIJCAI, pages 4940–4948,

    2023

  28. [2020]

    Dynamic many-objective molecular optimization: Unfolding com- plexity with objective decomposition and progressive opti- mization

    [Shinet al., 2024 ] Dong-Hee Shin, Young-Han Son, Deok- Joong Lee, Ji-Wung Han, and Tae-Eui Kam. Dynamic many-objective molecular optimization: Unfolding com- plexity with objective decomposition and progressive opti- mization. InProceedings of the Thirty-Third International J...

    2024

  29. [2021]

    Kipf and Max Welling

    [Kipf and Welling, 2017] Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. InInternational Conference on Learning Rep- resentations,

    2017

  30. [2022]

    Butler, Daniel W

    [Butleret al., 2018 ] Keith T. Butler, Daniel W. Davies, Hugh Cartwright, Olexandr Isayev, and Aron Walsh. Machine learning for molecular and materials science.Nature, page 547–555, Jul

    2018

  31. [2023]

    Ae-nerf: Augmenting event-based neural radiance fields for non-ideal conditions and larger scene,

    [Fenget al., 2025 ] Chaoran Feng, Wangbo Yu, Xinhua Cheng, Zhenyu Tang, Junwu Zhang, Li Yuan, and Yonghong Tian. Ae-nerf: Augmenting event-based neural radiance fields for non-ideal conditions and larger scene,

    2025

  32. [2024]

    [Fanget al., 2023 ] Y . Fang, Y . Jiang, L. Wei, Q. Ma, Z. Ren, Q. Yuan, and D. Q. Wei. Deepprosite: structure-aware pro- tein binding site prediction using esmfold and pretrained language model.Bioinformatics, 39(12):btad718, Dec

    2023

  33. [2025]

    Iqbal, M

    [Iqbalet al., 2021 ] J. Iqbal, M. V ogt, and J. Bajorath. Pre- diction of activity cliffs on the basis of images using con- volutional neural networks.Journal of Computer-Aided Molecular Design, pages 1–8,

    2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.