Pith. sign in

REVIEW 3 major objections 5 minor 3 cited by

Exploring Discrete Flow Matching for 3D De Novo Molecule Generation

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Discrete-state CTMC flow matching beats continuous approximations for 3D de novo molecule generation because discrete jumps eliminate atom-type assignment lag.

desk verdict Useful controlled benchmark and strong CTMC results, but the 'CTMC is inherently better' claim is undercut by asymmetric hyperparameter tuning, and the SOTA in the abstract needs a validity qualifier. read the letter →

arxiv 2411.16644 v1 pith:G3E5UE2G submitted 2024-11-25 cs.LG q-bio.BM

Ian Dunn , David R. Koes This is my paper · ORCID
classification cs.LGq-bio.BM
keywords discreteflowmatchingcontinuous-timeMarkovchains3Dmoleculegenerationdenovodrugdesignmolecularvaliditystructuralalerts
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 tries to establish which way of extending flow matching—a generative modeling framework originally designed for continuous data—should be used for the discrete labels (atom type, charge, bond order) in 3D molecule generation. Keeping the architecture, data, and training fixed, it finds that modeling these labels as discrete tokens that jump between a mask state and real values via a continuous-time Markov chain (CTMC) is far better than three continuous relaxations: 96.2% of generated molecules have valid valencies versus 20.4–69.5%, and 91.5% pass standard chemical validity checks on GEOM-Drugs. The paper explains the gap by measuring when an atom's final type appears in the sampled trajectory versus in the denoiser's prediction; continuous flows show a large lag, while CTMC flows do not. It also contributes quality metrics based on known problematic functional groups and ring systems absent from ChEMBL, which show all models, including the best, generate unusual motifs at rates well above the training data. A sympathetic reader would care because the results identify a concrete design choice—discrete jumps instead of smooth relaxations—and warn that standard validity metrics overstate molecular quality.

What carries the argument

The load-bearing mechanism is the masked CTMC flow, defined on a discrete state space in which every categorical variable is a token among the real atom types plus an extra mask token; the prior is all masked, and the conditional path at time $t$ holds the final token with probability $\kappa_t$ and the mask with probability $1-\kappa_t$. Sampling proceeds by categorical transitions whose rate matrix gives an unmasking rate that grows with the schedule's derivative and a stochastic remasking rate $\eta$, with $\eta=30$ and a temperature $\tau=0.05$ that sharpens the denoiser's token probabilities; these discrete jumps let the evolving sample immediately realize the denoiser's predicted token. The diagnostic that carries the explanation is the atom-type assignment time, defined as the first time an atom's final type appears in the sample trajectory and stays there, compared with the same time in the denoiser's output trajectory; the gap between these two times is large for continuous, simplex, and Dirichlet flows and small for CTMC. The paper also introduces the structural-alert and out-of-distribution-ring metrics as measurement machinery for molecule quality beyond valency.

What would settle it

Sweep the per-method hyperparameters—interpolant schedule shape, sampling temperature, and remasking rate—and recompute the GEOM-Drugs stable/valid percentages and the atom-type assignment-time CDFs for all four DFM variants. If a differently tuned continuous or simplex method erases most of CTMC's 26-point stability lead, or if CTMC's lead disappears when its eta and tau are re-tuned, the paper's attribution of the gap to discreteness is refuted; if the assignment-lag persists for continuous methods even after tuning, the mechanism survives.

Watch

Extended reading notes

Core claim

The central claim is that, for de novo 3D small-molecule generation under a fixed graph-network architecture, the way categorical features are treated in the flow matching formulation is the dominant factor in sample quality. The paper's FlowMol-CTMC model, which uses masked continuous-time Markov chain flows with low-temperature sampling, achieves 96.2% valency-stable and 91.5% RDKit-valid molecules on GEOM-Drugs, exceeding the diffusion baselines MiDi (85.1%/71.6%), JODO (90.7%/76.5%), and EQGAT-Diff (93.4%/86.1%) while using only 4.3M parameters. The proposed mechanism is atom-type assignment time: for continuous and simplex-constrained flows, the denoiser fixes an atom's final type early (78% at t=0.5 for the Continuous variant) but the evolving sample only approaches that one-hot vertex near t=1, creating a lag between decision and realization; CTMC flows jump to the unmasked predicted token immediately, shrinking the lag. A second, independent claim is that basic validity is not enough: using structural-alert lists and ChEMBL ring frequencies as metrics, the paper shows all tested models generate problematic and out-of-distribution functional groups at rates much higher than training data, with FlowMol-CTMC's validity advantage accompanied by more structural alerts per molecule than JODO (1.23 vs 0.84).

Load-bearing premise

The comparison assumes the timing curves and sampling settings chosen for each flow variant—linear for CTMC, cosine for the others, plus CTMC's temperature and remasking rate—are fair and near-optimal; if those settings happen to favor CTMC, the 26-point stability advantage and the assignment-lag explanation could be partly an artifact of tuning rather than of discrete flow matching itself.

Editorial extensions

If this is right

  • Changing only the categorical flow method—from a continuous one-hot flow to a masked CTMC flow—raises valency-stable molecules on GEOM-Drugs from 69.5% to 96.2%, so the discrete flow formulation, not the architecture, drives most of the validity gain.
  • FlowMol-CTMC reaches 91.5% RDKit-valid molecules with 4.3M parameters, outperforming MiDi, JODO, and EQGAT-Diff on stability and validity while using fewer parameters than all three.
  • Continuous and simplex-based flow matching methods show a measurable atom-type assignment lag—at t=0.5 the Continuous denoiser has fixed 78% of atom types while only 15% of sampled atoms occupy their final type—whereas CTMC flows jump to predicted states without this lag.
  • Valency-based validity is not a reliable proxy for medicinal-chemistry quality: all tested models emit structural alerts and ring systems unseen in ChEMBL at rates far above training data, with FlowMol-CTMC producing 1.23 structural alerts per molecule versus 0.84 for JODO.
  • Reporting only stability and validity can mislead; the proposed structural-alert and out-of-distribution-ring metrics should accompany future de novo generation results.

Reading between the lines

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

  • Inference: The assignment-lag mechanism likely transfers beyond molecules: continuous flows on one-hot or simplex embeddings of any categorical data—protein sequences, text tokens, or graph node labels—should show the same pattern, and measuring assignment-time CDFs in one such domain would test the transfer.
  • Inference: A testable extension is to hold the CTMC formulation fixed while varying only the interpolant schedule; if schedule shape, rather than discreteness, is a major driver of the gain, matching schedules across methods would shrink the gap and refine the paper's mechanism.
  • Inference: The structural-alert findings suggest a concrete training objective: add a penalty or reward for generated molecules containing known problematic functional groups or for ring systems absent from ChEMBL; if validity stays high while alert rates fall, the observed trade-off is a property of the training loss, not of CTMC flow matching itself.
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

3 major / 5 minor

Summary. The paper benchmarks four discrete flow matching (DFM) formulations for unconditional 3D de novo molecule generation within a fixed FlowMol architecture and training setup: a fully continuous baseline, SimplexFlow, Dirichlet Flows, and CTMC flows. On GEOM-Drugs (Table 1) the CTMC variant achieves substantially higher molecular stability and validity (96.2% and 91.5%, respectively) than the other DFM variants, and it also outperforms the external baselines MiDi, JODO, and EQGAT-Diff on these validity metrics with fewer parameters (Table 2). The authors propose two new quality metrics, structural alert rate and out-of-distribution ring rate, showing that all models generate problematic functional groups more often than the training data. They also propose an assignment-time analysis (Figure 2, Appendix C) to explain why CTMC flows outperform continuous/simplex-based flows, arguing that continuous representations introduce a time lag between denoiser predictions and their realization in the sampled trajectory.

Significance. If the comparison is fair, this is a valuable controlled benchmark: it holds architecture, dataset, and training procedure fixed while varying only the DFM formulation, and it provides open-source code and trained models. The use of 95% confidence intervals over 5x10,000 samples, the introduction of structural alert and out-of-distribution ring metrics, and the mechanistic assignment-time analysis are genuine contributions. However, the central causal claim that CTMC flows are inherently better at navigating discrete spaces, and the associated state-of-the-art framing, currently rest on a comparison in which the CTMC variant receives method-specific tuning (low-temperature sampling, stochasticity, and a linear schedule) that the other variants do not receive. The benchmark remains useful and the paper can be repaired, but the main claim needs deconfounding or careful restatement.

major comments (3)
  1. [§3.1, §A.4, §A.5] The statement in §3.1 that model variants "are distinguished only by the discrete flow matching method" is contradicted by the appendices. §A.4 shows CTMC uses low-temperature sampling with tau=0.05 and stochasticity eta=30, which the authors call "critical for model performance," and §A.5 gives CTMC a linear interpolant while all other variants use a cosine interpolant with modality-specific exponents taken from FlowMol [24]. No analogous temperature, stochasticity, or schedule search is reported for the Continuous, SimplexFlow, or Dirichlet variants. Because the 26-point stability gap in Table 1 and the assignment-time gap in Figure 2 are the main evidence for the claim that CTMC flows are "inherently better" at navigating discrete spaces, the comparison is confounded by asymmetric hyperparameter selection. I request either (a) a hyperparameter search for each variant over schedule, temperature, and stochasticity, with the final settings reported, or (b) a clear restatement of the conclusion as applying to the default settings tested, together with a sensitivity analysis showing how CTMC's stability/validity vary with tau and eta.
  2. [§4, Appendix C, Figure 2] The assignment-time explanation is not causally established. For the continuous variants, the cosine schedule kappa_t = 1 - cos^2(pi/2 t^nu) slows progress near t=1, so atom type vectors in g_t mechanically remain close to the prior or to mixed states until late in the trajectory. For CTMC, the linear schedule and the low-temperature denoiser (tau=0.05) make final assignments in both g_t and the denoiser occur earlier and more sharply. The reported difference in assignment-time lag may therefore be a property of the chosen schedules and decoding temperatures rather than an intrinsic property of CTMC flows. To support the mechanistic claim, the authors should compare assignment-time CDFs under matched schedules (e.g., CTMC with a cosine schedule, or continuous variants with a linear schedule and low-temperature softmax) and report whether the lag difference persists.
  3. [Abstract, §1, Table 2] The abstract's claim of "state of the art performance for 3D de novo design" is too broad and is not matched by Table 2. While FlowMol-CTMC has the best stability/validity among the compared baselines, its structural alert rate (1.23 per mol) is worse than all three baselines (0.84-1.06) and its OOD ring rate (0.28) is worse than JODO (0.21). The introduction and conclusion appropriately qualify the claim to "state of the art molecular validity," but the abstract does not. The claim should be reworded to be validity-specific and should explicitly acknowledge the functional-group trade-off. Additionally, Tables 1 and 2 report slightly different validity for the same model (91.5±0.3 vs 91.6±0.1); the reason for this discrepancy should be stated.
minor comments (5)
  1. [§A.4, Eq. (15) and surrounding text] The text after Eq. (15) says an atom has probability kappa_t of being in the masked state and probability 1-kappa_t of being in its final state, but Eq. (15) states p = kappa_t delta_{A1} + (1-kappa_t) delta_M, i.e., the probabilities are reversed in the prose. Please correct the prose.
  2. [Figure 2, Appendix C] The assignment-time CDFs in Figure 2 are computed from 100 sampled molecules, whereas Table 1 uses 5x10,000 samples; the figure would benefit from confidence intervals or at least a statement that the qualitative pattern is stable across repeated sampling runs and across random seeds.
  3. [§3.3 and Tables 1-2] The metric definitions for "Structural Alert Rate (per mol)" and "OOD Ring (per mol)" should specify whether multiple occurrences of the same functional group in one molecule are counted separately, and whether invalid or unsanitizable molecules are included in these rates; this affects interpretability of the per-mol averages.
  4. [Appendix E, Tables 3-4] The QM9 results are relegated to the appendix but show a different ranking among DFM variants (e.g., the Continuous variant reaches 96.8% stable, close to CTMC's 99.3%). A sentence in the main text explaining whether QM9 results support or qualify the GEOM-Drugs conclusions would help readers assess the generality of the main claim.
  5. [§2, Eq. (3)] The notation p_theta(x^1_i | x_t) is introduced with a superscript 1 that could be confused with an exponent; consider using p_theta(x^{(1)}_i | x_t) or defining the superscript explicitly as the endpoint time index to avoid ambiguity.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the paper is an empirical benchmark with independent external baselines; the only self-citations (FlowMol architecture and cosine schedule values) are not load-bearing for the central comparison.

full rationale

The paper does not derive its main result from its own assumptions by construction. FlowMol-CTMC's validity and stability scores are measured against external baselines (MiDi, JODO, EQGAT-Diff), and the compared DFM variants are implemented within the same architecture, so the benchmark has independent empirical content. The self-citation to FlowMol [24] supplies the model architecture and the cosine interpolant schedule values (Appendix A.5), but these are not used to define the outcome metrics and do not force the observed performance ordering. The CTMC sampling uses the authors' chosen low-temperature parameters (τ=0.05, η=30, Appendix A.4), which is a hyperparameter fairness concern rather than a circularity: the reported metrics are not algebraically determined by those parameters. The 'assignment time lag' analysis in Section 4 is an interpretation of measured trajectories, not a fitted parameter renamed as a prediction; although the CTMC transition rule directly samples from the denoiser, the paper does not use this equivalence to derive the validity results. No equation in the paper reduces to its own inputs, no uniqueness theorem is imported from the authors' prior work, and no known result is merely renamed. The appropriate finding is therefore no significant circularity, with the caveat that the per-method hyperparameter choices could affect the strength of the comparative claim, which is a correctness risk rather than a circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central empirical claim rests on evaluation assumptions (RDKit validity, MMFF energy, structural alerts, ChEMBL ring frequencies), architectural transferability of FlowMol to all DFM variants, and several hand-set sampling hyperparameters. No new physical entities are introduced.

free parameters (5)
  • CTMC sampling temperature tau = 0.05
    Appendix A.4 says low-temperature sampling is 'critical for model performance'; this is a hand-tuned sampling hyperparameter and the reported validity numbers depend on it.
  • CTMC stochasticity eta = 30
    Appendix A.4, eta controls the unmasking and remasking rate; it is tuned and used in the central comparison.
  • Dirichlet flow omega_max = 10
    Appendix A.3.2 sets omega_max=10 to connect the Dirichlet flow time to the interpolant schedule; this affects the Dirichlet ablation result.
  • Loss weights (eta_X, eta_A, eta_C, eta_E) = (3, 0.4, 1, 2)
    Appendix D adopts these weights from Vignac et al.; they are hand-chosen scalars that shape all model variants.
  • Cosine interpolant exponents nu per modality = values from FlowMol [24]
    Appendix A.5 uses a cosine interpolant for non-CTMC flows with nu values taken from the authors' prior FlowMol paper; this is a design choice that affects the continuous and simplex variants.
assumptions (4)
  • domain assumption The fixed FlowMol architecture is an adequate and equally well-matched denoiser for all four discrete flow matching variants.
    Section 3.1 and 3.2: variants differ only in DFM method. If the architecture is better matched to CTMC than to continuous flows, the performance gap is confounded.
  • domain assumption RDKit sanitization, valency stability, MMFF energies, Dundee and Glaxo structural alerts, and ChEMBL ring frequencies are meaningful measures of molecule quality.
    Section 3.3 defines validity and quality metrics; the paper's central evaluation and the new metrics rest on this assumption.
  • domain assumption The CTMC rate matrix formulation of Campbell et al. [29] is valid for molecule generation.
    Appendix A.4 adopts the rate matrix from Campbell et al. and the paper finds it better than the alternative from Gat et al.; this is an external modeling assumption.
  • domain assumption The equivariant optimal transport coupling for atomic coordinates from prior work is appropriate for molecule flow matching.
    Appendix A.1 uses the coupling from [31,46] to align prior and data coordinates; if this coupling is suboptimal, coordinate quality is affected.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploring Discrete Flow Matching for 3D De Novo Molecule Generation." pith.science (2026). https://pith.science/paper/G3E5UE2G

@misc{pith2026241116644,
  author       = {Pith},
  title        = {Pith review of: Exploring Discrete Flow Matching for 3D De Novo Molecule Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G3E5UE2G}},
  note         = {Machine review of arXiv:2411.16644}
}
read the original abstract

Deep generative models that produce novel molecular structures have the potential to facilitate chemical discovery. Flow matching is a recently proposed generative modeling framework that has achieved impressive performance on a variety of tasks including those on biomolecular structures. The seminal flow matching framework was developed only for continuous data. However, de novo molecular design tasks require generating discrete data such as atomic elements or sequences of amino acid residues. Several discrete flow matching methods have been proposed recently to address this gap. In this work we benchmark the performance of existing discrete flow matching methods for 3D de novo small molecule generation and provide explanations of their differing behavior. As a result we present FlowMol-CTMC, an open-source model that achieves state of the art performance for 3D de novo design with fewer learnable parameters than existing methods. Additionally, we propose the use of metrics that capture molecule quality beyond local chemical valency constraints and towards higher-order structural motifs. These metrics show that even though basic constraints are satisfied, the models tend to produce unusual and potentially problematic functional groups outside of the training data distribution. Code and trained models for reproducing this work are available at \url{https://github.com/dunni3/FlowMol}.

Figures

Figures reproduced from arXiv: 2411.16644 by the authors.

Figure 1
Figure 1. Overview Top: We adapt the flow matching framework for unconditional 3D molecule generation and explore the use of different discrete flow matching methods. This CTMC trajectory shows masked atoms in gray. Middle: Trajectory of the atom type vector for a single atom under SimplexFlow, a variant of continuous flow matching developed for categorical variables. Atom type flows lie on the probability simplex. Bottom: Tr… view at source ↗
Figure 2
Figure 2. Atom Type Assignment Times: Cumulative Density Functions (CDFs) of the time at which an atom is assigned its final atom type, for each DFM method tested. Green lines show the time of final atom type assignments in gt. Gold lines show the times when the final atom type is assigned in gˆ1(gt) (the predicted final molecule given the current molecule at time t). this phenomenon by measuring the time at which an atom acq… view at source ↗
Figure 3
Figure 3. Prior distributions used with support on the probability simplex [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FlowMol Architecture Top left: An input molecular graph gt is transformed into a predicted final molecular graph g1 by being passed through multiple molecule update blocks. Top right: A molecule update block uses NFU, NPU, and EFU sub-components to update all molecular…
Figure 5
Figure 5. Figure 5: Atom type trajectories and associated times of final type assignment. Each plot contains [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

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

  1. Boltzmann-Expected Molecular Design with Decoupled Annealing Flows

    stat.ML 2026-07 conditional novelty 7.0 of 10

    A two-flow simulated-annealing loop makes ensemble statistics—means, variances, and skewness of 3D properties—the objective of molecular graph design.

  2. FlowMol3: Flow Matching for 3D De Novo Small-Molecule Generation

    cs.LG 2025-08 conditional novelty 6.0 of 10

    Combining self-conditioning, fake atoms, and late-stage geometry distortion lets a compact flow-matching model generate nearly always valid 3D drug-like molecules and match training-data chemistry better than existing...

  3. An evaluation of unconditional 3D molecular generation methods

    physics.chem-ph 2025-05 conditional novelty 5.0 of 10

    An evaluation of five unconditional 3D molecular generators finds that standard saturated benchmarks overstate quality, and that post-processing changes the model ranking.

Reference graph

Works this paper leans on

59 extracted references · 19 canonical work pages · cited by 3 Pith papers

  1. [24]

    Mixed Continuous and Categorical Flow Matching for 3D De Novo Molecule Generation, April 2024

    Ian Dunn and David Ryan Koes. Mixed Continuous and Categorical Flow Matching for 3D De Novo Molecule Generation, April 2024. URL http://arxiv.org/abs/2404.19739. arXiv:2404.19739 [cs, q-bio]

    arXiv 2024

  2. [1]

    A dual diffusion model enables 3D molecule generation and lead optimization based on target pockets

    Lei Huang, Tingyang Xu, Yang Yu, Peilin Zhao, Xingjian Chen, Jing Han, Zhi Xie, Hailong Li, Wenge Zhong, Ka-Chun Wong, and Hengtong Zhang. A dual diffusion model enables 3D molecule generation and lead optimization based on target pockets. Nature Communications, 15 (1):2657, March 2024. ISSN 2041-1723. doi: 10.1038/s41467-024-46569-1. URL https:// www.nat...

    doi:10.1038/s41467-024-46569-1 2024

  3. [2]

    3D Equivariant Diffusion for Target-Aware Molecule Generation and Affinity Prediction, March

    Jiaqi Guan, Wesley Wei Qian, Xingang Peng, Yufeng Su, Jian Peng, and Jianzhu Ma. 3D Equivariant Diffusion for Target-Aware Molecule Generation and Affinity Prediction, March

    work page

  4. [3]

    Structure-based Drug Design with Equivariant Diffusion Models, June 2023

    Arne Schneuing, Yuanqi Du, Charles Harris, Arian Jamasb, Ilia Igashov, Weitao Du, Tom Blundell, Pietro Lió, Carla Gomes, Max Welling, Michael Bronstein, and Bruno Correia. Structure-based Drug Design with Equivariant Diffusion Models, June 2023. URL http: //arxiv.org/abs/2210.13695. arXiv:2210.13695 [cs, q-bio]

    arXiv 2023

  5. [4]

    Pocket2Mol: Efficient Molecular Sampling Based on 3D Protein Pockets, May 2022

    Xingang Peng, Shitong Luo, Jiaqi Guan, Qi Xie, Jian Peng, and Jianzhu Ma. Pocket2Mol: Efficient Molecular Sampling Based on 3D Protein Pockets, May 2022. URL http://arxiv. org/abs/2205.07249. arXiv:2205.07249 [cs, q-bio]

    arXiv 2022

  6. [5]

    Generating 3D Molecules for Target Protein Binding, May 2022

    Meng Liu, Youzhi Luo, Kanji Uchino, Koji Maruhashi, and Shuiwang Ji. Generating 3D Molecules for Target Protein Binding, May 2022. URL http://arxiv.org/abs/2204. 09410. arXiv:2204.09410 [cs, q-bio]

    arXiv 2022

  7. [6]

    Mathis, and Pietro Lio

    Jos Torge, Charles Harris, Simon V . Mathis, and Pietro Lio. DiffHopp: A Graph Diffusion Model for Novel Drug Design via Scaffold Hopping, August 2023. URL http://arxiv.org/ abs/2308.07416. arXiv:2308.07416 [q-bio]

    arXiv 2023

  8. [7]

    Equivariant 3D-conditional diffusion model for molecular linker design

    Ilia Igashov, Hannes Stärk, Clément Vignac, Arne Schneuing, Victor Garcia Satorras, Pascal Frossard, Max Welling, Michael Bronstein, and Bruno Correia. Equivariant 3D-conditional diffusion model for molecular linker design. Nature Machine Intelligence, pages 1–11, April

    work page

Show all 59 references
  1. [8]

    Accelerating Inference in Molecular Diffusion Models with La- tent Representations of Protein Structure

    Ian Dunn and David Koes. Accelerating Inference in Molecular Diffusion Models with La- tent Representations of Protein Structure. October 2023. URL https://openreview.net/ forum?id=Z4ia7s2tpV

    2023

  2. [9]

    Watson, David Juergens, Nathaniel R

    Joseph L. Watson, David Juergens, Nathaniel R. Bennett, Brian L. Trippe, Jason Yim, Helen E. Eisenach, Woody Ahern, Andrew J. Borst, Robert J. Ragotte, Lukas F. Milles, Basile I. M. Wicky, Nikita Hanikel, Samuel J. Pellock, Alexis Courbet, William Sheffler, Jue Wang, Preetham ...

    2023

  3. [10]

    Bennett, Joseph L

    Nathaniel R. Bennett, Joseph L. Watson, Robert J. Ragotte, Andrew J. Borst, Déjenaé L. See, Connor Weidle, Riti Biswas, Ellen L. Shrock, Philip J. Y . Leung, Buwei Huang, Inna Gore- shnik, Russell Ault, Kenneth D. Carr, Benedikt Singer, Cameron Criswell, Dionne Vafea- dos, Mar...

    2024 doi

  4. [11]

    Ingraham, Max Baranov, Zak Costello, Karl W

    John B. Ingraham, Max Baranov, Zak Costello, Karl W. Barber, Wujie Wang, Ahmed Ismail, Vincent Frappier, Dana M. Lord, Christopher Ng-Thow-Hing, Erik R. Van Vlack, Shan Tie, Vincent Xue, Sarah C. Cowles, Alan Leung, João V . Rodrigues, Claudio L. Morales-Perez, Alex M. Ayoub, ...

    2023

  5. [12]

    MatterGen: a generative model for inorganic materials design, January

    Claudio Zeni, Robert Pinsler, Daniel Zügner, Andrew Fowler, Matthew Horton, Xiang Fu, Sasha Shysheya, Jonathan Crabbé, Lixin Sun, Jake Smith, Bichlien Nguyen, Hannes Schulz, Sarah Lewis, Chin-Wei Huang, Ziheng Lu, Yichi Zhou, Han Yang, Hongxia Hao, Jielan Li, Ryota Tomioka, an...

  6. [13]

    Weiss, Niru Maheswaranathan, and Surya Ganguli

    Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep Unsupervised Learning using Nonequilibrium Thermodynamics, November 2015. URL http: //arxiv.org/abs/1503.03585. arXiv:1503.03585 [cond-mat, q-bio, stat]

    2015 arXiv

  7. [14]

    Denoising Diffusion Probabilistic Models, December

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising Diffusion Probabilistic Models, December

  8. [15]

    arXiv:2312.03687 [cond-mat]

    URL http://arxiv.org/abs/2312.03687. arXiv:2312.03687 [cond-mat]

    arXiv

  9. [16]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow Matching for Generative Modeling, February 2023. URL http://arxiv.org/abs/2210. 02747. arXiv:2210.02747 [cs, stat]

    2023 arXiv

  10. [17]

    Improving and generalizing flow-based generative models with minibatch optimal transport, July 2023

    Alexander Tong, Nikolay Malkin, Guillaume Huguet, Yanlei Zhang, Jarrid Rector-Brooks, Kilian Fatras, Guy Wolf, and Yoshua Bengio. Improving and generalizing flow-based generative models with minibatch optimal transport, July 2023. URL http://arxiv.org/abs/2302. 00482. arXiv:23...

    2023 arXiv

  11. [18]

    Albergo, Nicholas M

    Michael S. Albergo, Nicholas M. Boffi, and Eric Vanden-Eijnden. Stochastic Interpolants: A Unifying Framework for Flows and Diffusions, November 2023. URL http://arxiv.org/ abs/2303.08797. arXiv:2303.08797 [cond-mat]. 7

    2023 arXiv

  12. [19]

    Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole

    Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-Based Generative Modeling through Stochastic Differential Equations, February 2021. URL http://arxiv.org/abs/2011.13456. arXiv:2011.13456 [cs, stat]

    2021 arXiv

  13. [20]

    Albergo, Nicholas M

    Nanye Ma, Mark Goldstein, Michael S. Albergo, Nicholas M. Boffi, Eric Vanden-Eijnden, and Saining Xie. SiT: Exploring Flow and Diffusion-based Generative Models with Scal- able Interpolant Transformers, January 2024. URL http://arxiv.org/abs/2401.08740. arXiv:2401.08740 [cs]

    2024 arXiv

  14. [21]

    SE(3)-Stochastic Flow Matching for Protein Backbone Generation, April

    Avishek Joey Bose, Tara Akhound-Sadegh, Guillaume Huguet, Kilian Fatras, Jarrid Rector- Brooks, Cheng-Hao Liu, Andrei Cristian Nica, Maksym Korablyov, Michael Bronstein, and Alexander Tong. SE(3)-Stochastic Flow Matching for Protein Backbone Generation, April

  15. [22]

    Jason Yim, Andrew Campbell, Andrew Y . K. Foong, Michael Gastegger, José Jiménez-Luna, Sarah Lewis, Victor Garcia Satorras, Bastiaan S. Veeling, Regina Barzilay, Tommi Jaakkola, and Frank Noé. Fast protein backbone generation with SE(3) flow matching, October 2023. URL http://...

    2023 arXiv

  16. [23]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow, September 2022

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow, September 2022. URL http://arxiv.org/abs/ 2209.03003. arXiv:2209.03003 [cs]

    2022 arXiv

  17. [25]

    Fisher Flow Matching for Generative Modeling over Discrete Data, May

    Oscar Davis, Samuel Kessler, Mircea Petrache, ˙Ismail ˙Ilkan Ceylan, Michael Bronstein, and Avishek Joey Bose. Fisher Flow Matching for Generative Modeling over Discrete Data, May

  18. [26]

    arXiv:2310.02391 [cs]

    URL http://arxiv.org/abs/2310.02391. arXiv:2310.02391 [cs]

    arXiv

  19. [27]

    Categorical Flow Matching on Statistical Manifolds, May 2024

    Chaoran Cheng, Jiahan Li, Jian Peng, and Ge Liu. Categorical Flow Matching on Statistical Manifolds, May 2024. URL http://arxiv.org/abs/2405.16441. arXiv:2405.16441 [cs, stat]

    2024

  20. [28]

    Sequence-Augmented SE(3)-Flow Matching For Conditional Protein Backbone Generation, May 2024

    Guillaume Huguet, James Vuckovic, Kilian Fatras, Eric Thibodeau-Laufer, Pablo Lemos, Riashat Islam, Cheng-Hao Liu, Jarrid Rector-Brooks, Tara Akhound-Sadegh, Michael Bronstein, Alexander Tong, and Avishek Joey Bose. Sequence-Augmented SE(3)-Flow Matching For Conditional Protei...

    2024 arXiv

  21. [29]

    Genera- tive Flows on Discrete State-Spaces: Enabling Multimodal Flows with Applications to Protein Co-Design, June 2024

    Andrew Campbell, Jason Yim, Regina Barzilay, Tom Rainforth, and Tommi Jaakkola. Genera- tive Flows on Discrete State-Spaces: Enabling Multimodal Flows with Applications to Protein Co-Design, June 2024. URL http://arxiv.org/abs/2402.04997. arXiv:2402.04997 [cs, q-bio, stat]

    2024 arXiv

  22. [30]

    Itai Gat, Tal Remez, Neta Shaul, Felix Kreuk, Ricky T. Q. Chen, Gabriel Synnaeve, Yossi Adi, and Yaron Lipman. Discrete Flow Matching, July 2024. URL http://arxiv.org/abs/2407. 15595. arXiv:2407.15595 [cs]

    2024 arXiv

  23. [31]

    arXiv:2405.14664 [cs]

    URL http://arxiv.org/abs/2405.14664. arXiv:2405.14664 [cs]

    arXiv

  24. [32]

    Dirichlet Flow Matching with Applications to DNA Sequence Design, February 2024

    Hannes Stark, Bowen Jing, Chenyu Wang, Gabriele Corso, Bonnie Berger, Regina Barzilay, and Tommi Jaakkola. Dirichlet Flow Matching with Applications to DNA Sequence Design, February 2024. URL http://arxiv.org/abs/2402.05841. arXiv:2402.05841 [cs, q-bio]

    2024 arXiv

  25. [33]

    GEOM, energy-annotated molecular confor- mations for property prediction and molecular generation

    Simon Axelrod and Rafael Gómez-Bombarelli. GEOM, energy-annotated molecular confor- mations for property prediction and molecular generation. Scientific Data, 9(1):185, April

  26. [34]

    Variational Flow Matching for Graph Generation, June 2024

    Floor Eijkelboom, Grigory Bartosh, Christian Andersson Naesseth, Max Welling, and Jan- Willem van de Meent. Variational Flow Matching for Graph Generation, June 2024. URL http://arxiv.org/abs/2406.04843. arXiv:2406.04843 [cs, stat]

    2024 arXiv

  27. [35]

    Blum, and Jean-Louis Reymond

    Lars Ruddigkeit, Ruud van Deursen, Lorenz C. Blum, and Jean-Louis Reymond. Enumeration of 166 Billion Organic Small Molecules in the Chemical Universe Database GDB-17. Journal of Chemical Information and Modeling , 52(11):2864–2875, November 2012. ISSN 1549-

    2012

  28. [36]

    Dral, Matthias Rupp, and O

    Raghunathan Ramakrishnan, Pavlo O. Dral, Matthias Rupp, and O. Anatole von Lilienfeld. Quantum chemistry structures and properties of 134 kilo molecules. Scientific Data, 1(1): 140022, August 2014. ISSN 2052-4463. doi: 10.1038/sdata.2014.22. URL https://www. nature.com/article...

    2014 doi

  29. [37]

    Equivariant Flow Matching with Hybrid Probability Transport, December

    Yuxuan Song, Jingjing Gong, Minkai Xu, Ziyao Cao, Yanyan Lan, Stefano Ermon, Hao Zhou, and Wei-Ying Ma. Equivariant Flow Matching with Hybrid Probability Transport, December

  30. [38]

    arXiv:2312.07168 [cs]

    URL http://arxiv.org/abs/2312.07168. arXiv:2312.07168 [cs]

    arXiv

  31. [39]

    Soni, and Ron O

    Bowen Jing, Stephan Eismann, Pratham N. Soni, and Ron O. Dror. Equivariant Graph Neural Networks for 3D Macromolecular Structure, July 2021. URL http://arxiv.org/abs/2106. 03843. arXiv:2106.03843 [cs, q-bio]. 8

    2021 arXiv

  32. [40]

    Strategic Pooling of Compounds for High-Throughput Screening

    Mike Hann, Brian Hudson, Xiao Lewell, Rob Lifely, Luke Miller, and Nigel Ramsden. Strategic Pooling of Compounds for High-Throughput Screening. Journal of Chemical Information and Computer Sciences, 39(5):897–902, September 1999. ISSN 0095-2338. doi: 10.1021/ci990423o. URL htt...

    1999 doi

  33. [41]

    The ChEMBL Database in 2023: a drug discovery platform spanning multiple bioactivity data types and time periods

    Barbara Zdrazil, Eloy Felix, Fiona Hunter, Emma J Manners, James Blackshaw, Sybilla Corbett, Marleen de Veij, Harris Ioannidis, David Mendez Lopez, Juan F Mosquera, Maria Paula Magarinos, Nicolas Bosc, Ricardo Arcila, Tevfik Kizilören, Anna Gaulton, A Patrícia Bento, Melissa F...

    2023

  34. [42]

    MiDi: Mixed Graph and 3D Denoising Diffusion for Molecule Generation, June 2023

    Clement Vignac, Nagham Osman, Laura Toni, and Pascal Frossard. MiDi: Mixed Graph and 3D Denoising Diffusion for Molecule Generation, June 2023. URL http://arxiv.org/abs/ 2302.09048. arXiv:2302.09048 [cs]

    2023 arXiv

  35. [43]

    Learning Joint 2D & 3D Diffusion Models for Complete Molecule Generation, June 2023

    Han Huang, Leilei Sun, Bowen Du, and Weifeng Lv. Learning Joint 2D & 3D Diffusion Models for Complete Molecule Generation, June 2023. URL http://arxiv.org/abs/2305.12347. arXiv:2305.12347 [cs, q-bio]

    2023 arXiv

  36. [44]

    Navigating the Design Space of Equivariant Diffusion-Based Generative Models for De Novo 3D Molecule Generation, November 2023

    Tuan Le, Julian Cremer, Frank Noé, Djork-Arné Clevert, and Kristof Schütt. Navigating the Design Space of Equivariant Diffusion-Based Generative Models for De Novo 3D Molecule Generation, November 2023. URLhttp://arxiv.org/abs/2309.17296. arXiv:2309.17296 [cs]

    2023 arXiv

  37. [45]

    Efficient 3D Molecular Generation with Flow Matching and Scale Optimal Transport, June 2024

    Ross Irwin, Alessandro Tibo, Jon Paul Janet, and Simon Olsson. Efficient 3D Molecular Generation with Flow Matching and Scale Optimal Transport, June 2024. URLhttp://arxiv. org/abs/2406.07266. arXiv:2406.07266 [cs]

    2024 arXiv

  38. [46]

    URL http://www.rdkit.org/

    RDKit. URL http://www.rdkit.org/

  39. [47]

    Patrick W. Walters. Generative Molecular Design Isn’t As Easy As People Make It Look, May 2024. URL https://practicalcheminformatics.blogspot.com/2024/05/ generative-molecular-design-isnt-as.html

    2024

  40. [48]

    Lessons Learnt from Assembling Screening Libraries for Drug Dis- covery for Neglected Diseases

    Ruth Brenk, Alessandro Schipani, Daniel James, Agata Krasowski, Ian Hugh Gilbert, Julie Frear- son, and Paul Graham Wyatt. Lessons Learnt from Assembling Screening Libraries for Drug Dis- covery for Neglected Diseases. ChemMedChem, 3(3):435–444, 2008. ISSN 1860-7187. doi: 10. ...

    2008 doi

  41. [51]

    PatWalters/useful_rdkit_utils, September 2024

    Patrick Walters. PatWalters/useful_rdkit_utils, September 2024. URL https://github.com/ PatWalters/useful_rdkit_utils. original-date: 2021-12-31T00:24:33Z

    2024

  42. [55]

    Equivariant flow matching, November 2023

    Leon Klein, Andreas Krämer, and Frank Noé. Equivariant flow matching, November 2023. URL http://arxiv.org/abs/2306.15030. arXiv:2306.15030 [physics, stat]. 9

    2023 arXiv

  43. [56]

    interpolant schedule

    Minjie Wang, Da Zheng, Zihao Ye, Quan Gan, Mufei Li, Xiang Song, Jinjing Zhou, Chao Ma, Lingfan Yu, Yu Gai, Tianjun Xiao, Tong He, George Karypis, Jinyang Li, and Zheng Zhang. Deep Graph Library: A Graph-Centric, Highly-Performant Package for Graph Neural Networks, August 2020...

    2020 arXiv

  44. [57]

    The data distribution are one-hot vectors indicating the type of each atom

    = QN i=1 N (ai 0|0, I). The data distribution are one-hot vectors indicating the type of each atom. The conditional probability path is the same used for atomic coordinate flows (8). The learned vector field for sampling pt(At) is of the same form as well (9). However, we use ...

  45. [58]

    marginal-simplex

    is to define a continuous representation of categorical variables, and then construct a flow matching model where flows are constrained to this representation. The continuous representation chosen for a d-categorical variable is the d-dimensional probability simplex S d: S d =...

  46. [59]

    Note that we also use an abbreviated notation here where δA1 (ai) = δai 1 (ai). Atom type trajectories are sampled by iterative sampling of transition distributions that factorize over atoms: p(At+∆t|At) = NY i=1 pi(ai t+∆t|At) (16) The per-atom transition distributions are ca...

  47. [2020]

    arXiv:2006.11239 [cs, stat]

    URL http://arxiv.org/abs/2006.11239. arXiv:2006.11239 [cs, stat]

    2006 arXiv

  48. [2022]

    doi: 10.1038/s41597-022-01288-4

    ISSN 2052-4463. doi: 10.1038/s41597-022-01288-4. URL https://www.nature. com/articles/s41597-022-01288-4 . Publisher: Nature Publishing Group

    doi

  49. [2023]

    arXiv:2303.03543 [cs, q-bio]

    URL http://arxiv.org/abs/2303.03543. arXiv:2303.03543 [cs, q-bio]

    arXiv

  50. [2024]

    doi: 10.1038/s42256-024-00815-9

    ISSN 2522-5839. doi: 10.1038/s42256-024-00815-9. URL https://www.nature. com/articles/s42256-024-00815-9 . Publisher: Nature Publishing Group. 6

    doi

  51. [9596]

    URL https://doi.org/10.1021/ci300415d

    doi: 10.1021/ci300415d. URL https://doi.org/10.1021/ci300415d. Publisher: American Chemical Society

    doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.