Pith. sign in

REVIEW 5 major objections 6 minor 58 references

Fused Bayesian Flow Networks for Dual-Target Molecular Design

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read FusedBFN claims that dual-target 3D molecular design is best done by fusing the distributions of two single-target Bayesian flow networks in a shared continuous parameter space, and reports better dual-target docking than drift-based…

desk verdict A legitimate parameter-space fusion paper with a real novelty and an untested PoE assumption; deserves peer review despite a decoding inconsistency. read the letter →

arxiv 2608.01007 v1 pith:SCJRUWXM submitted 2026-08-02 cs.LG

classification cs.LG
keywords dual-targetdrugdesignBayesianflownetworksproductofexpertsstructure-based3Dmoleculegenerationparameter-spacefusionpocketalignmentpolypharmacology
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 is trying to establish that dual-target 3D molecular design can be achieved by fusing two single-target Bayesian flow networks in the continuous parameter space where BFN updates live, rather than by training on dual-target complex data or by adding a drift term to a diffusion trajectory. It claims that modeling the dual-target sender distribution as a product of the two single-target sender distributions lets both binding-pocket contexts participate continuously in generation, and that the resulting fused flow keeps a closed form for atom coordinates and atom types. Using a frozen pretrained target-aware BFN backbone together with chemically aware or prior-free pocket alignment, the paper reports that the generated molecules dock well to both targets simultaneously and retain favorable drug-like properties. If the claim holds, it offers a zero-fine-tuning route from single-target generative models to polypharmacological compound design that beats the previous drift-based dual-target diffusion approach on the benchmark.

What carries the argument

The central object is the fused Bayesian flow distribution over the parameter-space variables: the mean coordinate $\mu$ and the categorical atom-type parameters $\theta^v$. It is built from a product-of-experts sender, $p_S(y|m_1,m_2,Tp_1,p_2;\alpha) \propto p_S(y|m_1,Tp_1;\alpha)\,p_S(y|m_2,p_2;\alpha)$, which for coordinates makes the fused sender a Gaussian centered at the average of the two single-target coordinates with half the variance, and for atom types a Gaussian whose mean is the average of the two one-hot projections. The load-bearing property is the additivity of sender accuracies for the fused flow, proved for continuous and discrete variables, which keeps the Bayesian update closed-form; combined with an SE(3)-equivariant shared network and a zeroed center of mass, the generative process stays rotation- and translation-equivariant. The two pockets are brought into a common frame either by chemically aware alignment of probe-ligand poses, which weights heavy atoms more heavily, or by aligning pocket-surface point clouds with RANSAC followed by ICP.

What would settle it

Take the same frozen backbone, the same alignment, and the same 12,917 target-pair benchmark, but replace the product-of-experts sender with a mixture or a convex average of the two single-target senders. If dual high affinity and Max Vina Dock do not get worse, or improve, then the product form is not the component doing the work; a second decisive test is to train on a set of real dual-target complexes and compare against the frozen-backbone variant to see whether the capacity limitation flagged in its limitations section actually costs affinity.

Watch

Extended reading notes

Core claim

FusedBFN's central claim is that dual-target generation is best done as distribution fusion in BFN parameter space: conditioned on two aligned pockets, the sender is the product of the two single-target senders, and because sender accuracies stay additive under this product, the fused Bayesian flow has a closed form for both continuous coordinates and discrete atom types. The paper shows that sampling with a frozen SE(3)-equivariant network as shared backbone produces molecules whose Vina docking is strong on both targets, with Max Vina Dock -8.02 on average and Dual High Affinity 57.8%, both better than the strongest drift-based baseline DualDiff (-7.60 and 51.2%), with the differences statistically significant. It also reports lower strain energy and lower docked-pose RMSD across atom-number ranges, meaning the generated molecules bind the two targets with more consistent conformations, together with a roughly tenfold sampling speedup over DualDiff.

Load-bearing premise

The load-bearing premise is that the dual-target conditional distribution is well approximated by the product of the two single-target conditionals, so product-of-experts fusion optimizes the right objective; the paper does not compare this fusion operator against alternatives, and its own limitations section notes that a frozen single-target backbone may not fully capture dual-target binding patterns.

Editorial extensions

If this is right

  • If FusedBFN is correct, structure-based dual-target design no longer depends on scarce dual-target complex data: a frozen single-target BFN backbone can be repurposed by fusing its flows for two pockets, at least on this benchmark.
  • Parameter-space fusion is the operative ingredient: the paper's ablation shows that fusing parameters (P-Fused) beats averaging the network's molecular estimates in sample space (S-Fused) on dual-target affinity and molecular properties.
  • Alignment quality directly controls dual-target performance: weighting heavy atoms during probe-ligand alignment improves over equal-atom alignment, and the prior-free surface alignment (surface-atom extraction plus RANSAC plus ICP) comes close to prior-based alignment while removing the need for probe ligands.
  • The fused-flow formulation carries an efficiency dividend: sampling 10 molecules for a target pair takes about 113.7 seconds with FusedBFN versus about 1128.4 seconds with DualDiff, which makes large-scale dual-target screening more practical.
  • Generated molecules are more conformationally consistent across the two binding sites, with lower strain energy and lower docked-pose RMSD, so they are closer to plausible dual-binding poses even before any force-field refinement.

Reading between the lines

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

  • Editorial inference: the product-of-experts fusion behaves like a logical AND over binding-site compatibility; a testable consequence is that FusedBFN should produce low-mass or empty generations when the two pockets demand incompatible scaffolds, since the product distribution then has little probability mass.
  • Editorial inference: the same additivity argument should extend to more than two targets by multiplying $k$ sender distributions; whether quality degrades gracefully with $k$, and whether the variance should be divided by $k$, are open empirical questions the paper does not test.
  • Editorial inference: because the reported gains are measured by docking after generation, part of the improvement could come from the improved pocket alignment rather than from the fusion itself; an experiment with deliberately misaligned pockets would separate the two contributions.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes FusedBFN, a zero-shot extension of the single-target Bayesian flow network MolCRAFT to dual-target structure-based drug design. Dual-target generation is formulated as product-of-experts fusion of two single-target sender distributions in the continuous parameter space of a BFN, and closed-form fused Bayesian updates are derived for continuous coordinates and discrete atom types. The authors also introduce a chemically aware ligand-prior alignment method and a prior-free pocket-surface alignment method, and evaluate on the 12,917-pair dual-target benchmark against TargetDiff, MolCRAFT, CompDiff, and DualDiff, reporting improvements in Max Vina Dock, Dual High Affinity, strain energy, and clash metrics while preserving molecular properties.

Significance. If the central mechanism is sound, this is a useful advance: it shows that a pretrained single-target SBDD model can be repurposed for dual-target generation without training or fine-tuning, with explicit closed-form Bayesian updates (Propositions 1 and 2) and an SE(3)-equivariance argument (Proposition 3). The empirical gains over DualDiff in Table 1 (Max Vina Dock -8.02 vs -7.60; Dual High Affinity 57.8% vs 51.2%) and the supporting ablations (parameter- vs sample-space fusion, alignment variants, probe generator choice) are substantive. However, the paper's load-bearing novelty, the product-of-experts fusion in Eq. (4), is not validated against alternative fusion operators, and the final decoding step is inconsistent with the PoE construction; the affinity evaluation also relies entirely on re-docking rather than direct pose quality. These issues mean the reported improvements cannot yet be attributed specifically to the claimed fusion mechanism.

major comments (5)
  1. [§4.1, Eq. (4)] Equation (4) introduces the product-of-experts sender distribution as the starting point of fusion, but this is an unvalidated modeling assumption rather than a derived consequence of dual-target binding. For coordinates, Eq. (7) forces the fused source to be the arithmetic mean of the two target-specific coordinates with doubled precision; for discrete types, Eq. (13) replaces the two target-specific sender means by their average in the Gaussian sender space. The manuscript states that this form is 'inspired by multimodal conditional image synthesis' but reports no comparison with alternative fusion operators such as a mixture sender, a weighted product with learned target reliability, or a learned gating mechanism. Because Propositions 1 and 2 and the entire fused update depend on this factorization, the current experiments do not establish that the PoE form is the mechanism behind the reported gains.
  2. [§4.1, Algorithm 1 (lines 17-18)] The final decoding step is not consistent with the PoE construction used during the updates. For continuous coordinates, sampling from the average of the two coordinate estimates matches the Gaussian PoE mean in Eq. (7), but for atom types, line 18 samples from the arithmetic average of the two output distributions, whereas the product-of-experts update in Eq. (13) and the fused flow in Eq. (18) require a product of the two experts. For categorical distributions these two operations differ. Thus the final generated molecule is not actually drawn from the fused distribution the paper claims to define. The authors should either implement PoE decoding, for example by sampling from the normalized product of the two output distributions, or explicitly state that final discrete decoding uses a different fusion operator and justify that choice.
  3. [§5.1, Table 1 and Fig. 2] The affinity evaluation rests entirely on Vina Dock, a re-docking procedure that optimizes the ligand pose inside each pocket independently. This can compensate for poorly placed generated atoms and does not demonstrate that the generated molecule can adopt a single conformation compatible with both pockets simultaneously. Figure 2 reports RMSD between two separately docked poses, which is not a direct measure of a dual-target co-complex. I recommend adding direct-pose metrics such as Vina Score and Vina Min evaluated on the generated coordinates, as the authors themselves use in Table 5 for the single-target setting, and, if feasible, a co-docking or combined-pocket evaluation. Without such evidence, the headline claim of simultaneous dual-target binding is stronger than what the experiments establish.
  4. [§5.2 and Table 3] The comparison between FusedBFN and DualDiff in Table 1 is not controlled with respect to the base generative model: FusedBFN starts from MolCRAFT, while DualDiff starts from a TargetDiff-style diffusion model, so the reported advantage could come from the stronger single-target backbone rather than from parameter-space fusion. Table 3's P-Fused versus S-Fused comparison is informative, but it compares two fusion-in-update strategies within FusedBFN; it is not a drift-based dual-target variant of MolCRAFT. To support the statement in §5.2 that 'information fusion in the continuous parameter space is more advantageous than drift in the mixed continuous-discrete sample space,' the paper should add a drift-based MolCRAFT adaptation or, equivalently, apply the proposed PoE fusion to the TargetDiff backbone.
  5. [Section E] The authors explicitly acknowledge that extending a pretrained single-target model to the dual-target setting 'may limit its capacity to fully capture the binding patterns between molecules and dual-target simultaneously.' Since zero-shot use of a frozen backbone is a central design choice and the final method is evaluated only in that configuration, the manuscript should analyze the impact of this limitation, for example by comparing with a fine-tuned or lightly adapted backbone or by reporting failure modes. The acknowledgment is to the authors' credit, but as written it qualifies the central claim that FusedBFN generates molecules with strong dual-target binding.
minor comments (6)
  1. [Reproducibility] The manuscript contains no code or data availability statement; providing the implementation and evaluation scripts would substantially help reproduction of the 12,917-pair benchmark and the alignment pipeline.
  2. [Table 1] The reference-ligand row displays six affinity numbers without clear per-metric Avg./Med. grouping; the column layout should be fixed to match the header structure.
  3. [Figure 2] Figure 2 would benefit from error bars or per-bin sample counts, since the current plot shows only means and makes it hard to assess the robustness of the RMSD differences.
  4. [Section D.3] The paired t-tests report p-values only; please also report effect sizes and confidence intervals, and state whether each target pair contributes a single averaged score or per-molecule scores.
  5. [Algorithm 1] The symbol t is used both as the loop-derived scalar in line 10 and as continuous time in the update function; renaming the former to t_i and adding parentheses in line 17 would remove confusion.
  6. [Section C.4] The chemically aware weighting is fixed at lambda_heavy=1 and lambda_light=0.1 without a sensitivity analysis; given this is advertised as a contribution, varying these weights in an ablation would strengthen the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the fused flow is derived from an explicit PoE ansatz and an external pretrained backbone, with no fitted parameter or self-citation chain carrying the measured claims.

full rationale

The central derivation starts from Eq. 4, where the dual-target sender distribution is explicitly modeled as the product of the two single-target sender distributions. This is a stated modeling assumption borrowed from multimodal image synthesis (refs. [44,45]), not a quantity fitted to the evaluation metrics. The subsequent algebra (Eqs. 5-18) is a direct consequence of this assumption together with the Gaussian/categorical sender forms inherited from BFN; it does not re-insert the experimentally measured binding affinities. No parameter in FusedBFN is fitted to Vina Dock, Dual High Affinity, QED, SA, or Diversity: the alignment weights (lambda_heavy=1, lambda_light=0.1), noise schedule (beta_1=1.5, sigma_1=0.03), and 100 sampling steps are hand-set, and the backbone is the externally pretrained MolCRAFT model. The dual-target benchmark and the DualDiff/CompDiff baselines come from the external work [26], so the reported improvement is not supported by a self-citation chain. The paper's own limitation statement in Section E concedes that the frozen single-target backbone may not fully capture dual-target binding patterns, which is a correctness caveat rather than a circular reduction. The only internal inconsistency is that Algorithm 1 (line 18) decodes atom types from an arithmetic average of the two output distributions instead of the product used in Eq. 4; but this is a mismatch between the stated fusion rule and the sampling implementation, not an instance of a prediction being equivalent to its input by construction. Therefore, no circular step satisfying the required evidentiary standard is present.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the PoE fusion assumption, the transferability of a frozen single-target backbone, alignment validity, standard BFN additivity, SE(3)-equivariance, and docking as affinity proxy. The paper introduces seven hand-set hyperparameters that influence the experiments. No invented entities are introduced.

free parameters (7)
  • lambda_heavy = 1
    Hand-chosen weight for heavy atoms in the chemically aware prior alignment (Eq. 19, Section C.4). No sensitivity analysis is provided, and it directly affects the pocket transformation used in the central experiments.
  • lambda_light = 0.1
    Hand-chosen weight for light atoms in the same alignment (Eq. 19, Section C.4). Used to downweight light atoms; the value is an assumption, not fitted.
  • sampling_steps = 100
    Number of Bayesian flow steps during generation (Section C.3). Chosen by the authors; affects generation quality and runtime.
  • ransac_iterations = 1000
    Iterations for RANSAC coarse alignment of pocket surfaces (Section C.4). Standard choice, but it is a hand-set parameter of the proposed alignment pipeline.
  • icp_iterations = 50
    Maximum iterations for ICP refinement in surface alignment (Section C.4). Affects alignment convergence and final generation.
  • noise_schedule_beta1 = 1.5
    Noise schedule for discrete atom types (Section C.3). Inherited from MolCRAFT-style settings but stated as the paper's experimental configuration.
  • noise_schedule_sigma1 = 0.03
    Noise schedule for continuous coordinates (Section C.3). Same status as beta1.
assumptions (6)
  • ad hoc to paper Product-of-experts sender distribution (Eq. 4): pS(y|m1,m2,Tp1,p2;α) ∝ p(y|m1,Tp1;α) p(y|m2,p2;α)
    Central modeling choice, motivated by multimodal conditional image synthesis [44,45], not derived from binding physics. If the per-target senders are not conditionally independent given the molecule, the product form is an approximation. Appears in Section 4.1.
  • domain assumption Pretrained single-target BFN generalizes to dual-target without retraining
    The frozen MolCRAFT backbone is assumed to represent dual-target binding modes when driven by the fused flow. The authors' own Limitations section (Section E) states this may limit capacity. This assumption is load-bearing for the claim of training-free extension.
  • domain assumption A single rigid alignment maps both pockets into a common frame suitable for one ligand pose
    Both alignment methods (prior-based and surface-based) assume that a shared coordinate frame exists in which a single molecule can satisfy both binding sites. Section 4.2 and Limitations note dependence on alignment quality.
  • standard math Additivity of sender accuracies in standard BFN
    Propositions 1 and 2 rely on the additivity property established for BFNs by Graves et al. [27]. This is background mathematics, not proved in this paper.
  • domain assumption The backbone network Ψ is SE(3)-equivariant
    Proposition 3 requires Ψ to be an SE(3)-equivariant GNN, inherited from MolCRAFT. The proof of equivariance of the fused process depends on this property.
  • domain assumption AutoDock Vina re-docking scores are a valid proxy for dual-target binding affinity
    The central evaluation treats Vina Dock as the measure of binding. This is standard in the field but imperfect; the paper does not validate with physical assays or alternative scoring functions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fused Bayesian Flow Networks for Dual-Target Molecular Design." pith.science (2026). https://pith.science/paper/SCJRUWXM

@misc{pith2026260801007,
  author       = {Pith},
  title        = {Pith review of: Fused Bayesian Flow Networks for Dual-Target Molecular Design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SCJRUWXM}},
  note         = {Machine review of arXiv:2608.01007}
}
read the original abstract

Dual-target drug design aims to generate 3D molecules that can simultaneously interact with two target proteins, offering a promising route for discovering polypharmacological compounds against complex diseases. While recent generative models have shown encouraging performance in single-target drug design, existing dual-target approaches either focus on sequence generation or introduce an additional predictive drift term into the diffusion-based generative trajectory, which limits their ability to fully integrate feature information from both targets. We propose FusedBFN, a fused Bayesian flow network (BFN) for dual-target molecular design. FusedBFN formulates dual-target generation as distribution fusion in a unified continuous parameter space and employs a product-of-experts formulation to incorporate dual-target information throughout the generative process. To address the scarcity of dual-target structural data, we leverage a pretrained target-aware BFN model as the shared backbone. We further introduce a chemically aware prior-based alignment method and a prior-free pocket alignment strategy to construct aligned dual-target contexts. Extensive experiments demonstrate that FusedBFN generates molecules with strong binding affinity toward dual targets while maintaining favorable molecular properties.

Figures

Figures reproduced from arXiv: 2608.01007 by the authors.

Figure 1
Figure 1. Overview of FusedBFN for dual-target molecular generation. We first aligns two protein [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. RMSD distributions of docked poses on dual targets across different atom-number intervals. Lower values indicate better cross-target pose con￾sistency. Binding Affinity and Molecular Properties We evaluate all methods under the dual-target setting, with the results reported in [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Visualizations of reference ligands and molecules generated by different methods for the [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Visualization of more reference ligands and molecules generated by MolCRAFT, DualDiff [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 47 canonical work pages

  1. [1]

    Protein modeling and structure-based drug design,

    G. Klebe, “Protein modeling and structure-based drug design,” inDrug Design: From Structure and Mode-of-Action to Rational Design Concepts. Springer, 2025, pp. 309–321

  2. [2]

    Geometric deep learning methods and applications in 3d structure-based drug design,

    Q. Bai, T. Xu, J. Huang, and H. Perez-Sanchez, “Geometric deep learning methods and applications in 3d structure-based drug design,”Drug Discovery Today, vol. 29, no. 7, p. 104024, 2024

  3. [3]

    From magic bullets to designed multiple ligands,

    R. Morphy, C. Kay, and Z. Rankovic, “From magic bullets to designed multiple ligands,”Drug discovery today, vol. 9, no. 15, pp. 641–651, 2004

  4. [4]

    Multitarget drug discovery and polypharmacology,

    M. L. Bolognesi and A. Cavalli, “Multitarget drug discovery and polypharmacology,” pp. 1190–1192, 2016

  5. [5]

    Computational polypharmacology: a new paradigm for drug discovery,

    R. Chaudhari, Z. Tan, B. Huang, and S. Zhang, “Computational polypharmacology: a new paradigm for drug discovery,”Expert opinion on drug discovery, vol. 12, no. 3, pp. 279–291, 2017

  6. [6]

    De novo generation of multi-target compounds using deep generative chemistry,

    B. P. Munson, M. Chen, A. Bogosian, J. F. Kreisberg, K. Licon, R. Abagyan, B. M. Kuenzi, and T. Ideker, “De novo generation of multi-target compounds using deep generative chemistry,” Nature Communications, vol. 15, no. 1, p. 3636, 2024

  7. [7]

    Architecture of the human regulatory network derived from encode data,

    M. B. Gerstein, A. Kundaje, M. Hariharan, S. G. Landt, K.-K. Yan, C. Cheng, X. J. Mu, E. Khurana, J. Rozowsky, R. Alexanderet al., “Architecture of the human regulatory network derived from encode data,”Nature, vol. 489, no. 7414, pp. 91–100, 2012

  8. [8]

    Ligbuilder v3: a multi-target de novo drug design approach,

    Y . Yuan, J. Pei, and L. Lai, “Ligbuilder v3: a multi-target de novo drug design approach,” Frontiers in chemistry, vol. 8, p. 142, 2020

Show all 58 references
  1. [9]

    Polypharmacology in a single drug: multitarget drugs,

    M. L Bolognesi, “Polypharmacology in a single drug: multitarget drugs,”Current medicinal chemistry, vol. 20, no. 13, pp. 1639–1645, 2013

  2. [10]

    A perspec- tive on multi-target drug discovery and design for complex diseases,

    R. R. Ramsay, M. R. Popovic-Nikolic, K. Nikolic, E. Uliassi, and M. L. Bolognesi, “A perspec- tive on multi-target drug discovery and design for complex diseases,”Clinical and translational medicine, vol. 7, no. 1, p. 3, 2018

  3. [11]

    Combination therapeutics in complex diseases,

    B. He, C. Lu, G. Zheng, X. He, M. Wang, G. Chen, G. Zhang, and A. Lu, “Combination therapeutics in complex diseases,”Journal of cellular and molecular medicine, vol. 20, no. 12, pp. 2231–2240, 2016

  4. [12]

    Generation of dual-target compounds using a transformer chemical language model,

    S. Srinivasan and J. Bajorath, “Generation of dual-target compounds using a transformer chemical language model,”Cell Reports Physical Science, vol. 5, no. 11, 2024

  5. [13]

    Therapeutic strategies of dual-target small molecules to overcome drug resistance in cancer therapy,

    J. Ye, J. Wu, and B. Liu, “Therapeutic strategies of dual-target small molecules to overcome drug resistance in cancer therapy,”Biochimica et Biophysica Acta (BBA)-Reviews on Cancer, vol. 1878, no. 3, p. 188866, 2023

  6. [14]

    Beyond purinergic signaling: Dual ligand approaches to multitarget drug discovery,

    B. Francucci, A. Smirnov, A. Spinaci, M. Buccioni, and G. Marucci, “Beyond purinergic signaling: Dual ligand approaches to multitarget drug discovery,”Drug Discovery Today, p. 104457, 2025

  7. [15]

    A 3d generative model for structure-based drug design,

    S. Luo, J. Guan, J. Ma, and J. Peng, “A 3d generative model for structure-based drug design,” Advances in Neural Information Processing Systems, vol. 34, pp. 6229–6239, 2021. 10

  8. [16]

    Pocket2mol: Efficient molecular sampling based on 3d protein pockets,

    X. Peng, S. Luo, J. Guan, Q. Xie, J. Peng, and J. Ma, “Pocket2mol: Efficient molecular sampling based on 3d protein pockets,” inInternational Conference on Machine Learning. PMLR, 2022, pp. 17 644–17 655

  9. [17]

    3d equivariant diffusion for target- aware molecule generation and affinity prediction,

    J. Guan, W. W. Qian, X. Peng, Y . Su, J. Peng, and J. Ma, “3d equivariant diffusion for target- aware molecule generation and affinity prediction,”arXiv preprint arXiv:2303.03543, 2023

  10. [18]

    Decompdiff: diffusion models with decomposed priors for structure-based drug design,

    J. Guan, X. Zhou, Y . Yang, Y . Bao, J. Peng, J. Ma, Q. Liu, L. Wang, and Q. Gu, “Decompdiff: diffusion models with decomposed priors for structure-based drug design,”arXiv preprint arXiv:2403.07902, 2024

  11. [19]

    Kgdiff: towards explainable target-aware molecule generation with knowledge guidance,

    H. Qian, W. Huang, S. Tu, and L. Xu, “Kgdiff: towards explainable target-aware molecule generation with knowledge guidance,”Briefings in Bioinformatics, vol. 25, no. 1, p. bbad435, 2024

  12. [20]

    Aligning target-aware molecule diffusion models with exact energy optimization,

    S. Gu, M. Xu, A. Powers, W. Nie, T. Geffner, K. Kreis, J. Leskovec, A. Vahdat, and S. Ermon, “Aligning target-aware molecule diffusion models with exact energy optimization,”Advances in Neural Information Processing Systems, vol. 37, pp. 44 040–44 063, 2024

  13. [21]

    Protein-ligand interaction prior for binding-aware 3d molecule diffusion models,

    Z. Huang, L. Yang, X. Zhou, Z. Zhang, W. Zhang, X. Zheng, J. Chen, Y . Wang, B. Cui, and W. Yang, “Protein-ligand interaction prior for binding-aware 3d molecule diffusion models,” in The Twelfth International Conference on Learning Representations, 2024

  14. [22]

    Molcraft: Structure-based drug design in continuous parameter space,

    Y . Qu, K. Qiu, Y . Song, J. Gong, J. Han, M. Zheng, H. Zhou, and W.-Y . Ma, “Molcraft: Structure-based drug design in continuous parameter space,”arXiv preprint arXiv:2404.12141, 2024

  15. [23]

    Multi-objective structure-based drug design using causal discovery,

    J. Zhou, D. Zhao, H. Qian, S. Tu, and L. Xu, “Multi-objective structure-based drug design using causal discovery,”IEEE Transactions on Computational Biology and Bioinformatics, 2025

  16. [24]

    Prior-guided flow matching for target-aware molecule design with learnable atom number,

    J. Zhou, H. Qian, S. Tu, and L. Xu, “Prior-guided flow matching for target-aware molecule design with learnable atom number,”arXiv preprint arXiv:2509.01486, 2025

  17. [25]

    Automated design of multi-target ligands by generative deep learning,

    L. Isigkeit, T. Hörmann, E. Schallmayer, K. Scholz, F. F. Lillich, J. H. Ehrler, B. Hufnagel, J. Büchner, J. A. Marschner, J. Pabelet al., “Automated design of multi-target ligands by generative deep learning,”Nature Communications, vol. 15, no. 1, p. 7946, 2024

  18. [26]

    Reprogramming pretrained target-specific diffusion models for dual-target drug design,

    X. Zhou, J. Guan, Y . Zhang, X. Peng, L. Wang, and J. Ma, “Reprogramming pretrained target-specific diffusion models for dual-target drug design,”Advances in Neural Information Processing Systems, vol. 37, pp. 87 255–87 281, 2024

  19. [27]

    Bayesian flow networks,

    A. Graves, R. K. Srivastava, T. Atkinson, and F. Gomez, “Bayesian flow networks,”arXiv preprint arXiv:2308.07037, 2023

  20. [28]

    Training products of experts by minimizing contrastive divergence,

    G. E. Hinton, “Training products of experts by minimizing contrastive divergence,”Neural computation, vol. 14, no. 8, pp. 1771–1800, 2002

  21. [29]

    Generating 3d molecules conditional on receptor binding sites with deep generative models,

    M. Ragoza, T. Masuda, and D. R. Koes, “Generating 3d molecules conditional on receptor binding sites with deep generative models,”Chemical science, vol. 13, no. 9, pp. 2701–2713, 2022

  22. [30]

    Generating 3d molecules for target protein binding,

    M. Liu, Y . Luo, K. Uchino, K. Maruhashi, and S. Ji, “Generating 3d molecules for target protein binding,” inInternational Conference on Machine Learning (ICML), 2022

  23. [31]

    Molecule generation for target protein binding with structural motifs,

    Z. Zhang, Y . Min, S. Zheng, and Q. Liu, “Molecule generation for target protein binding with structural motifs,” inThe eleventh international conference on learning representations, 2023

  24. [32]

    Learning subpocket prototypes for generalizable structure-based drug design,

    Z. Zhang and Q. Liu, “Learning subpocket prototypes for generalizable structure-based drug design,” inInternational Conference on Machine Learning. PMLR, 2023, pp. 41 382–41 398

  25. [33]

    Flexsbdd: Structure-based drug design with flexible protein modeling,

    Z. Zhang, M. Wang, and Q. Liu, “Flexsbdd: Structure-based drug design with flexible protein modeling,”Advances in Neural Information Processing Systems, vol. 37, pp. 53 918–53 944, 2024. 11

  26. [34]

    Empower structure-based molecule optimization with gradient guided bayesian flow networks,

    K. Qiu, Y . Song, J. Yu, H. Ma, Z. Cao, Z. Zhang, Y . Wu, M. Zheng, H. Zhou, and W.-Y . Ma, “Empower structure-based molecule optimization with gradient guided bayesian flow networks,” inInternational Conference on Machine Learning. PMLR, 2025, pp. 50 645–50 671

  27. [35]

    Structure-aware dual-target drug design through collaborative learning of pharmacophore combination and molecular simulation,

    S. Chen, J. Xie, R. Ye, D. D. Xu, and Y . Yang, “Structure-aware dual-target drug design through collaborative learning of pharmacophore combination and molecular simulation,”Chemical science, vol. 15, no. 27, pp. 10 366–10 380, 2024

  28. [36]

    Unified generative modeling of 3d molecules with bayesian flow networks,

    Y . Song, J. Gong, H. Zhou, M. Zheng, J. Liu, and W.-Y . Ma, “Unified generative modeling of 3d molecules with bayesian flow networks,” inThe Twelfth International Conference on Learning Representations, 2024

  29. [37]

    Image-to-image bayesian flow networks with structurally informative priors,

    H. Dou, J. Du, X. Jiang, H. Li, W. Yao, and Y . Deng, “Image-to-image bayesian flow networks with structurally informative priors,”IEEE Transactions on Image Processing, 2025

  30. [38]

    Target-guided bayesian flow networks for quantitatively constrained cad generation,

    W. Zheng, C. Sun, W. Zhang, J. Lv, and X. Liu, “Target-guided bayesian flow networks for quantitatively constrained cad generation,” inProceedings of the 33rd ACM International Conference on Multimedia, 2025, pp. 3330–3339

  31. [39]

    Knowledge guided bayesian flow network for cad sequence generation

    L. Penglin, W. Zhang, X. Liu, and W. Zheng, “Knowledge guided bayesian flow network for cad sequence generation.”

  32. [40]

    Controllable molecule generation by sampling in continuous parameter space

    W. Zhang, Y . Sun, X. Liuet al., “Controllable molecule generation by sampling in continuous parameter space.”

  33. [41]

    Protein sequence modelling with bayesian flow networks,

    T. Atkinson, T. D. Barrett, S. Cameron, B. Guloglu, M. Greenig, C. B. Tan, L. Robinson, A. Graves, L. Copoiu, and A. Laterre, “Protein sequence modelling with bayesian flow networks,” Nature Communications, vol. 16, no. 1, p. 3197, 2025

  34. [42]

    Empower structure-based molecule optimization with gradient guided bayesian flow networks,

    K. Qiu, Y . Song, J. Yu, H. Ma, Z. Cao, Z. Zhang, Y . Wu, M. Zheng, H. Zhou, and W.-Y . Ma, “Empower structure-based molecule optimization with gradient guided bayesian flow networks,” arXiv preprint arXiv:2411.13280, 2024

  35. [43]

    Full-atom peptide design via riemannian-euclidean bayesian flow networks,

    H. Qian, S. Tu, and L. Xu, “Full-atom peptide design via riemannian-euclidean bayesian flow networks,”arXiv preprint arXiv:2511.14516, 2025

  36. [44]

    Multimodal variational autoencoders for semi-supervised learning: In defense of product-of-experts,

    S. Kutuzova, O. Krause, D. McCloskey, M. Nielsen, and C. Igel, “Multimodal variational autoencoders for semi-supervised learning: In defense of product-of-experts,”arXiv preprint arXiv:2101.07240, 2021

  37. [45]

    Multimodal conditional image synthesis with product-of-experts gans,

    X. Huang, A. Mallya, T.-C. Wang, and M.-Y . Liu, “Multimodal conditional image synthesis with product-of-experts gans,” inEuropean conference on computer vision. Springer, 2022, pp. 91–109

  38. [46]

    Equivariant diffusion for molecule generation in 3d,

    E. Hoogeboom, V . G. Satorras, C. Vignac, and M. Welling, “Equivariant diffusion for molecule generation in 3d,” inInternational conference on machine learning. PMLR, 2022, pp. 8867– 8887

  39. [47]

    Equivariant flows: exact likelihood generative learning for symmetric densities,

    J. Köhler, L. Klein, and F. Noé, “Equivariant flows: exact likelihood generative learning for symmetric densities,” inInternational conference on machine learning. PMLR, 2020, pp. 5361–5370

  40. [48]

    E (n) equivariant normalizing flows,

    V . Garcia Satorras, E. Hoogeboom, F. Fuchs, I. Posner, and M. Welling, “E (n) equivariant normalizing flows,”Advances in Neural Information Processing Systems, vol. 34, pp. 4181– 4192, 2021

  41. [49]

    Geodiff: A geometric diffusion model for molecular conformation generation,

    M. Xu, L. Yu, Y . Song, C. Shi, S. Ermon, and J. Tang, “Geodiff: A geometric diffusion model for molecular conformation generation,”arXiv preprint arXiv:2203.02923, 2022

  42. [50]

    Identification of cavities on protein sur- face using multiple computational approaches for drug binding site prediction,

    Z. Zhang, Y . Li, B. Lin, M. Schroeder, and B. Huang, “Identification of cavities on protein sur- face using multiple computational approaches for drug binding site prediction,”Bioinformatics, vol. 27, no. 15, pp. 2083–2088, 2011. 12

  43. [51]

    P2rank: machine learning based tool for rapid and accurate prediction of ligand binding sites from protein structure,

    R. Krivák and D. Hoksza, “P2rank: machine learning based tool for rapid and accurate prediction of ligand binding sites from protein structure,”Journal of cheminformatics, vol. 10, no. 1, p. 39, 2018

  44. [52]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,

    M. A. Fischler and R. C. Bolles, “Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,”Communications of the ACM, vol. 24, no. 6, pp. 381–395, 1981

  45. [53]

    Method for registration of 3-d shapes,

    P. J. Besl and N. D. McKay, “Method for registration of 3-d shapes,” inSensor fusion IV: control paradigms and data structures, vol. 1611. Spie, 1992, pp. 586–606

  46. [54]

    Posecheck: Generative models for 3d structure-based drug design produce unrealistic poses,

    C. Harris, K. Didi, A. R. Jamasb, C. K. Joshi, S. V . Mathis, P. Lio, and T. L. Blundell, “Posecheck: Generative models for 3d structure-based drug design produce unrealistic poses,” 2023

  47. [55]

    Autodock vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading,

    O. Trott and A. J. Olson, “Autodock vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading,”Journal of computational chemistry, vol. 31, no. 2, pp. 455–461, 2010

  48. [56]

    Quantifying the chemical beauty of drugs,

    G. R. Bickerton, G. V . Paolini, J. Besnard, S. Muresan, and A. L. Hopkins, “Quantifying the chemical beauty of drugs,”Nature chemistry, vol. 4, no. 2, pp. 90–98, 2012

  49. [57]

    Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions,

    P. Ertl and A. Schuffenhauer, “Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions,”Journal of cheminformatics, vol. 1, pp. 1–11, 2009. 13 A Algorithm The sampling procedure of FusedBFN are summarized belo...

  50. [58]

    Avg.”, “T-Avg

    to estimate the mean, trimmed mean (i.e., averaging that removes the top 10% and bottom 10% of values before calculating) and median (denoted as “Avg.”, “T-Avg.” and “Med.” respectively) of affinity-related metrics. Vina Score evaluates binding affinity based on the generated ...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.