Pith. sign in

REVIEW 4 major objections 6 minor 54 references

AbFlowNet: Optimizing Antibody-Antigen Binding Energy via Diffusion-GFlowNet Fusion

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read AbFlowNet shows that a diffusion model for antibody CDR design can be trained to optimize binding energy by adding a GFlowNet Trajectory Balance objective whose terminal reward is a precomputed Rosetta energy score.

desk verdict Solid new application of GFlowNet TB to diffusion-based CDR design, with honest reporting; the energy improvements are real only for the Rosetta surrogate, not for true binding affinity. read the letter →

arxiv 2505.12358 v1 pith:X4ICHQOD submitted 2025-05-18 cs.LG cs.AI

Abrar Rahman Abir , Haz Sameen Shahgir , Md Rownok Zahan Ratul , Md Toki Tahmid , Greg Ver Steeg , Yue Dong This is my paper · ORCID
classification cs.LGcs.AI
keywords antibodydesignCDRdiffusionmodelsGFlowNetTrajectoryBalancebindingenergyRosettaproteingeneration
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

AbFlowNet aims to show that antibody CDR design can be made to optimize binding energy directly, without the expensive online reinforcement-learning stage used by prior methods. It fuses a diffusion model for CDR sequence and structure with a GFlowNet, treating each denoising step as a state and using the Trajectory Balance objective to propagate a terminal reward—the Rosetta-computed binding energy of the reference CDR, precomputed once per training complex—back through the whole trajectory. On the RAbD test set, averaged over all six CDR regions, AbFlowNet improves amino acid recovery by 3.06%, backbone RMSD by 20.40%, and the fraction of CDRs with better-than-reference binding energy by 3.60% relative to the same diffusion baseline at the same training budget. At 100 samples per complex, its Top-1 total energy and binding energy improve by 24.8% and 38.1% over the baseline, without using test-set structures during training. A sympathetic reading is that the paper thereby unifies reconstruction and reward optimization in one training procedure.

What carries the argument

The load-bearing mechanism is the GFlowNet Trajectory Balance objective applied to the diffusion trajectory. In its standard form it enforces $Z_\theta \prod_t P_F(s_t|s_{t-1}) = R(s_n)\prod_t P_B(s_{t-1}|s_t)$, where $Z_\theta$ is a learned estimate of the initial-state flow, $P_F$ is the diffusion model's denoiser, $P_B$ is the fixed forward noising process, and $R(s_n)$ is the exponential Rosetta binding-energy reward. Applied to CDR design, each partially denoised CDR is a state and the per-residue transition probabilities factor across amino-acid type, 3D coordinate, and SO(3) orientation. The objective propagates the terminal energy reward back through all 100 denoising steps while the diffusion reconstruction losses preserve fidelity to the training distribution.

What would settle it

Retrain AbFlowNet under the identical pipeline but with the terminal reward replaced by experimental in vitro binding-affinity measurements (or by an independently validated energy function) for the training complexes, then compare generated CDRs against DiffAb on both Rosetta energy and measured affinity. If the reported gains vanish when the reward is decoupled from the evaluation metric, the binding-energy improvement is an artifact of reward–metric coupling.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that a diffusion model for antibody CDR design can be trained to optimize binding energy by adding a GFlowNet Trajectory Balance term to the standard denoising losses. The GFlowNet state is a partially denoised CDR; the forward edge flow is the product of the learned denoising probabilities over all CDR residues, the backward edge flow is the product of the fixed noising probabilities, and the terminal state's flow is $\exp(-\alpha \cdot \text{BindingEnergy}(S_0))$ precomputed from Rosetta InterfaceAnalyzer for each training CDR. Enforcing parity between forward and backward trajectory flows makes the sparse terminal reward influence every denoising step. The paper reports that, with identical hyperparameters and gradient-update counts, this joint objective outperforms the base diffusion model on every averaged reconstruction and binding metric and is competitive with an online DPO-based method while using no test-set sampling and far less compute.

Load-bearing premise

The load-bearing premise is that the precomputed Rosetta binding-energy score used as the training reward is a useful signal for real binding affinity; since the same estimator also defines the reported energy improvements, the training loop and the evaluation metric are the same quantity, so if that estimator is unreliable the binding-energy gains may not reflect true affinity.

Editorial extensions

If this is right

  • Binding-energy optimization no longer requires sampling new CDRs and scoring them during training; a single precomputed energy per training complex suffices, cutting compute by orders of magnitude relative to online RL.
  • Because the reward is computed only on training complexes, the method avoids the test-set pseudo-labeling used by some RL baselines, reducing data-leakage concerns.
  • Joint optimization preserves reconstruction quality: unlike DPO-based post-training, which lowers amino acid recovery and worsens RMSD, AbFlowNet improves both relative to the base diffusion model.
  • If more accurate or experimental binding data became available for training complexes, the same pipeline could use them directly, since energies are needed only once per CDR rather than iteratively for generated CDRs.
  • The Top-1 energy gains at a 100-sample budget suggest that larger sampling budgets would likely produce further improvements, as the reference baseline itself improves from 480 to 211 kcal/mol when its budget grows from 100 to 2,528 samples.

Reading between the lines

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

  • Because the training reward and the reported energy metrics are both computed with Rosetta InterfaceAnalyzer, the measured binding-energy improvements may partly reflect overfitting to that estimator rather than true affinity gains; a test with experimental binding data or a different energy function would separate the two.
  • The same diffusion-as-GFlowNet framing could transfer to other sparse-reward protein generation tasks—side-chain packing, protein–protein docking, or enzyme design—whenever a terminal score can be precomputed once per training example.
  • The training schedule that activates the Trajectory Balance term only for the final 5,000 steps suggests the reward acts as a refinement signal on an already-trained generative model; ablating the schedule's length and position might reveal whether earlier reward injection helps or destabilizes reconstruction.
  • The paper's own appendix reports that Detailed Balance was infeasible because side-chain packing makes per-state energy evaluation too slow; a fast neural surrogate for Rosetta's interface energy would make local, per-step reward objectives viable and could be a concrete next step.
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

4 major / 6 minor

Summary. The paper proposes AbFlowNet, a CDR design model that adds a GFlowNet Trajectory Balance objective to the DiffAb diffusion backbone, using a precomputed terminal reward exp(-alpha * BindingEnergy(S0)) based on Rosetta InterfaceAnalyzer energies of the training complexes. Experiments on RAbD and a 19-complex DiffAb test set report average improvements of +3.06% amino acid recovery, +20.40% RMSD, and +3.60% IMP over DiffAb, together with Top-1 CDR total-energy and binding-energy reductions of 24.8% and 38.1% at a sampling budget of N=100. The authors release code and model weights and emphasize that AbFlowNet avoids online RL and test-set pseudo-labeling.

Significance. If the structural results are reliable, the article makes a modest but useful contribution: it demonstrates that a trajectory-balance term can be appended to a diffusion antibody model without sacrificing reconstruction quality, while improving a Rosetta-based energy surrogate. The release of code and weights, the matched gradient-update comparison with DiffAb, and the explicit reporting of hyperparameter sweeps are strengths. However, the binding-energy headline is not supported as a claim about true affinity, because the training reward and the evaluation metric are the same Rosetta InterfaceAnalyzer quantity, and the manuscript itself identifies this estimator as unreliable. The AAR and RMSD gains are independent of that issue and support the reconstruction claims, but they do not validate the energy claim.

major comments (4)
  1. [Section 5, Section 6.1, Eq. (17)] The terminal reward in Eq. (17) is R(S0) = exp(-alpha * BindingEnergy(S0)) with BindingEnergy computed by Rosetta InterfaceAnalyzer, and the Top-1 CDR Etotal and CDR-Ag Delta G metrics in Table 1 are the same InterfaceAnalyzer outputs. Since Section 2 and Appendix E state that InterfaceAnalyzer is an unreliable estimator of binding energy, the reported 24.8% and 38.1% improvements show that AbFlowNet optimizes the surrogate it was trained on, not that it produces tighter-binding antibodies. I request a concrete guard: evaluate on the subset of complexes with experimental affinity data, or use an independent estimator, or at minimum reframe the headline as 'Rosetta InterfaceAnalyzer energy' and remove the affinity implications from the abstract and conclusion.
  2. [Section A.2, Figure 4] The trajectory-balance weight w = 5e-6 in Eq. (18) is selected by evaluating on the RAbD dataset, which is the same test set used for all main results in Table 1 and Table 2. This makes the central comparison vulnerable to test-set peeking during hyperparameter selection. Please choose w on a held-out validation split derived from the SAbDab training set, then report final evaluation on RAbD and the DiffAb test set only for the fixed configuration.
  3. [Section 4.2, Eqs. (13)-(17), Section 7] Equation (17) writes log[ Z_theta * prod_t p(St | St-1; theta) / (R(S0) * prod_t q(St-1 | St)) ]^2, but Eqs. (13)-(16) define q(St | St-1) as the forward noising process and p(St-1 | St) as the reverse denoising process, so the TB-loss arguments appear reversed and p(St | St-1; theta) is not defined by Eq. (14). In addition, Section 7 states that backpropagation is performed for only one random timestep, which deviates from the full Trajectory Balance objective and makes the implemented loss an approximation. Because the GFlowNet objective is the core novelty, please correct the indexing and analyze or quantify the effect of the single-step gradient approximation on reward propagation.
  4. [Table 1, Section 6.1] The claim that AbFlowNet is 'competitive with AbDPO' is based on an apples-to-oranges comparison: AbDPO is evaluated at N=2,528 while AbFlowNet is evaluated at N=100, and Top-1 metrics are explicitly sampling-budget-sensitive, as the paper itself notes for DiffAb. The comparison against DiffAb at N=100 is fair, but the AbDPO competitiveness claim should either be removed, or supported by running AbFlowNet at the same N=2,528 budget on at least a subset of the test complexes.
minor comments (6)
  1. [Abstract] The phrase 'online Reinforcement Learning (RL) pipelines rely heavily on unreliable binding energy estimators' is a grammatical fragment; also, 'ABFlowNet' in the abstract should be 'AbFlowNet' for consistency.
  2. [Figure 3 caption] The caption says the generated CDRs were 'selected the one with the highest Delta G', but selecting the highest Delta G would pick the least favorable binding-energy value; this should read 'lowest Delta G' if the displayed values are the best energies.
  3. [Table 4] The column headers 'CDR+Etotal' and 'CDR-Ag+Delta G' are confusing, and the percent reductions 89.5% and 149.7% are unexplained; please clarify the normalization and the direction of improvement.
  4. [Section 4.2] The sentence 'we compute all forward q(St-1 | St) and backward probabilities p(St | St-1; theta)' uses notation reversed with respect to Eqs. (13)-(16); please make the time indexing consistent throughout Section 4.
  5. [Section 6.2, Table 2] The text says AbFlowNet 'outperforms DiffAb in all three metrics', but CDR-H3 IMP (14.38 vs 12.65) and CDR-L3 IMP (36.98 vs 34.70) both decrease; the claim should be qualified as 'on average across the six CDR regions'.
  6. [References] References [2] and [3] are duplicate entries for the same RAbD paper; please deduplicate.

Circularity Check

1 steps flagged · score 6.0 of 10

Binding-energy reward and evaluation are the same Rosetta InterfaceAnalyzer quantity; structural metrics remain independent.

  1. self definitional [Section 4.2, Eq. (17); Section 5 (Metrics); Section 7 (Discussion)]
    "Following Kim et al. [22], we precompute the reward R(S0) = exp(−α·BindingEnergy(S0)) for each CDR S0 in the training dataset and enforce the TB objective ... The binding energy is calculated by InterfaceAnalyzer in the Rosetta software package [2, 6] ... we used InterfaceAnalyzer to estimate the energies for our training set."

    Equation (17) defines the training reward as a monotone function of Rosetta InterfaceAnalyzer's BindingEnergy computed on training CDRs. Section 5 defines the evaluation metrics IMP, Top-1 CDR Etotal, and CDR-Ag ΔG with the sentence 'The binding energy is calculated by InterfaceAnalyzer'. Thus the quantity maximized during training and the quantity reported as 'binding energy improvement' are the same estimator applied to training and test CDRs respectively. The headline energy reductions (24.8% Etotal, 38.1% ΔG, +3.60% IMP) therefore verify that the model optimizes the training surrogate, not that it produces higher-affinity antibodies.

full rationale

The non-energy claims are self-contained: AAR and RMSD compare generated sequences and Cα coordinates against reference CDRs and do not depend on the reward function. The GFlowNet/Trajectory-Balance derivation is also mathematically coherent given the diffusion formulation. However, the paper's central energy claim is circular in validation: Eq. (17) trains AbFlowNet to increase exp(-alpha * BindingEnergy(S0)) with BindingEnergy supplied by Rosetta InterfaceAnalyzer, and Section 5's IMP and Top-1 energy metrics are the same InterfaceAnalyzer quantity. Consequently, the reported energy improvements show better optimization of the training surrogate, not independent evidence of improved binding affinity. The paper partially discloses this by calling InterfaceAnalyzer 'an unreliable estimator of binding energy' in Appendix E and noting only 'moderate correlation with the real binding energy' in Section 2, but the disclosed limitation does not remove the self-referential nature of the energy claim. No load-bearing self-citation chain or imported uniqueness theorem is present. Score 6 reflects partial circularity: the energy headline reduces to the training-reward/evaluation identity, while the structural reconstruction claims remain independent.

Assumptions & free parameters 4 free parameters · 3 assumptions · 2 invented entities

The central claim rests on the standard diffusion/CDR-design machinery, on a Rosetta estimator as reward, and on an empirically chosen TB loss weight and training schedule. There is no fitted parameter that is claimed to be a prediction; the closest concern is the unreported reward temperature alpha. The invented entities are mathematical devices from the GFlowNet framework, not empirical postulations.

free parameters (4)
  • TB loss weight w = 5e-6 in the main results
    Chosen by a parameter sweep reported in Appendix A.2; the final w is not derived from any principle, and the range of w tested spans 5e-5 to 1e-7.
  • Reward temperature alpha = not reported
    The reward definition R(S0) = exp(-alpha * BindingEnergy(S0)) includes an alpha that the paper never reports; this parameter determines how sharply the reward concentrates on low-energy CDRs and is not accounted for anywhere.
  • Number of TB training steps (5,000) = 5,000
    The final 5,000 steps of TB training, after 195,000 reconstruction-only steps, were chosen empirically; the appendix experiments show that 10,000 TB steps can hurt AAR and RMSD, so the schedule is a tuned value.
  • Training/test CDR-H3 sequence identity threshold = 50%
    The clustering and deduplication threshold of 50% sequence identity, taken from DiffAb, is a dataset-construction choice that controls how similar a test CDR can be to a training CDR; it affects both difficulty and leakage.
assumptions (3)
  • domain assumption Rosetta InterfaceAnalyzer scores are suitable as both a training reward and an evaluation metric for de novo CDRs.
    The entire energy-optimization claim assumes that lower InterfaceAnalyzer energy corresponds to better binding. The paper cites prior work (Vreven et al.) showing weak correlation with real binding energies, and Appendix D describes their own surrogate with Pearson coefficient 0.21, so the estimator is acknowledged to be unreliable.
  • domain assumption Framing each diffusion denoising step as a GFlowNet state with the TB objective does not introduce harmful bias in the diffusion trajectory.
    The TB loss couples the full forward and backward trajectory, and the authors backpropagate through only one timestep while computing the loss for all 100 steps; this makes the realized objective an approximation of TB and it is not shown that this approximation preserves the TB fixed point or beneficial credit assignment.
  • domain assumption The RAbD and DiffAb test sets, after 50% CDR-H3 sequence identity filtering, do not overlap with the training set in ways that bias the metric.
    The filtering relies on CDR-H3 sequence identity only; structural redundancy or framework-region similarity could still cause partial leakage, and this is not analyzed beyond the stated threshold.
invented entities (2)
  • Z_theta, the learned initial-state flow parameter
    purpose: A single learned scalar that approximates the total flow from the initial state in the Trajectory Balance equation (Eq. 17).
    This is a standard GFlowNet device to make the TB objective tractable; it is a learned parameter, not an observable, and the paper gives no specific accountability for its final value.
  • The GFlowNet state space over all partially denoised CDRs
    purpose: A mathematical re-labeling of noise levels as GFlowNet states.
    This is a bookkeeping device from the framework, not a new physical entity or conserved quantity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AbFlowNet: Optimizing Antibody-Antigen Binding Energy via Diffusion-GFlowNet Fusion." pith.science (2026). https://pith.science/paper/X4ICHQOD

@misc{pith2026250512358,
  author       = {Pith},
  title        = {Pith review of: AbFlowNet: Optimizing Antibody-Antigen Binding Energy via Diffusion-GFlowNet Fusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X4ICHQOD}},
  note         = {Machine review of arXiv:2505.12358}
}
read the original abstract

Complementarity Determining Regions (CDRs) are critical segments of an antibody that facilitate binding to specific antigens. Current computational methods for CDR design utilize reconstruction losses and do not jointly optimize binding energy, a crucial metric for antibody efficacy. Rather, binding energy optimization is done through computationally expensive Online Reinforcement Learning (RL) pipelines rely heavily on unreliable binding energy estimators. In this paper, we propose AbFlowNet, a novel generative framework that integrates GFlowNet with Diffusion models. By framing each diffusion step as a state in the GFlowNet framework, AbFlowNet jointly optimizes standard diffusion losses and binding energy by directly incorporating energy signals into the training process, thereby unifying diffusion and reward optimization in a single procedure. Experimental results show that AbFlowNet outperforms the base diffusion model by 3.06% in amino acid recovery, 20.40% in geometric reconstruction (RMSD), and 3.60% in binding energy improvement ratio. ABFlowNet also decreases Top-1 total energy and binding energy errors by 24.8% and 38.1% without pseudo-labeling the test dataset or using computationally expensive online RL regimes.

Figures

Figures reproduced from arXiv: 2505.12358 by the authors.

Figure 1
Figure 1. AbFlowNet reframes the diffusion process as a GFlowNet where each partially denoised [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Antibody-antigen complex. As shown in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. De novo Generated and Reference CDR-H3s for 5MES complex. For DiffAb and AbFlowNet, we generated 100 CDRs and selected the one with the highest ∆G. apoptosis. The Fab antibody serves to stabilize the complex, allowing researchers to resolve at at 2.24 Å resolution by X-ray diffraction [20] [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Hyperparameter search for TB loss weight [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Training Diffusion+GFlowNet models with different training steps on the RAbD dataset [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 31 canonical work pages

  1. [1]

    Determining kinetics and affinities of protein interactions using a parallel real-time label-free biosensor, the octet

    Yasmina Abdiche, Dan Malashock, Alanna Pinkerton, and Jaume Pons. Determining kinetics and affinities of protein interactions using a parallel real-time label-free biosensor, the octet. Analytical biochemistry, 377(2):209–217, 2008

    work page 2008

  2. [3]

    Rosettaantibodydesign (rabd): A general framework for computational antibody design.PLoS computational biology, 14(4): e1006112, 2018

    Jared Adolf-Bryfogle, Oleks Kalyuzhniy, Michael Kubitz, Brian D Weitzner, Xiaozhen Hu, Yumiko Adachi, William R Schief, and Roland L Dunbrack Jr. Rosettaantibodydesign (rabd): A general framework for computational antibody design.PLoS computational biology, 14(4): e1006112, 2018

    work page 2018

  3. [4]

    Flow network based generative models for non-iterative diverse candidate generation.Advances in Neural Information Processing Systems, 34:27381–27394, 2021

    Emmanuel Bengio, Moksh Jain, Maksym Korablyov, Doina Precup, and Yoshua Bengio. Flow network based generative models for non-iterative diverse candidate generation.Advances in Neural Information Processing Systems, 34:27381–27394, 2021

    work page 2021

  4. [5]

    Isothermal titration calorimetry of ion-coupled membrane transporters.Methods, 76:171–182, 2015

    Olga Boudker and SeCheol Oh. Isothermal titration calorimetry of ion-coupled membrane transporters.Methods, 76:171–182, 2015

    work page 2015

  5. [6]

    Pyrosetta: a script-based interface for implementing molecular modeling algorithms using rosetta.Bioinformatics, 26(5):689–691, 2010

    Sidhartha Chaudhury, Sergey Lyskov, and Jeffrey J Gray. Pyrosetta: a script-based interface for implementing molecular modeling algorithms using rosetta.Bioinformatics, 26(5):689–691, 2010

    2010

  6. [7]

    Prediction of absolute protein-protein binding free energy by a super learner model

    Elton Chaves, Emmanuel Mhrous, Manasses Nascimento-Filho, Carlos Cruz, Matheus Ferraz, and Roberto Lins. Prediction of absolute protein-protein binding free energy by a super learner model. 2023

    work page 2023

  7. [8]

    Noise contrastive alignment of language models with explicit rewards.arXiv preprint arXiv:2402.05369, 2024

    Huayu Chen, Guande He, Lifan Yuan, Ganqu Cui, Hang Su, and Jun Zhu. Noise contrastive alignment of language models with explicit rewards.arXiv preprint arXiv:2402.05369, 2024

    arXiv 2024

  8. [9]

    Lau, and Victor Ovchinnikov

    Simone Conti, Edmond Y . Lau, and Victor Ovchinnikov. On the rapid calculation of binding affinities for antigen and antibody design and affinity maturation simulations.Antibodies, 11 (3), 2022. ISSN 2073-4468. doi: 10.3390/antib11030051. URL https://www.mdpi.com/ 2073-4468/11/3/51

    work page doi:10.3390/antib11030051 2022

Show all 54 references
  1. [10]

    Investigat- ing data contamination in modern benchmarks for large language models.arXiv preprint arXiv:2311.09783, 2023

    Chunyuan Deng, Yilun Zhao, Xiangru Tang, Mark Gerstein, and Arman Cohan. Investigat- ing data contamination in modern benchmarks for large language models.arXiv preprint arXiv:2311.09783, 2023

    2023 arXiv

  2. [11]

    Sabdab: the structural antibody database.Nucleic acids research, 42(D1):D1140–D1146, 2014

    James Dunbar, Konrad Krawczyk, Jinwoo Leem, Terry Baker, Angelika Fuchs, Guy Georges, Jiye Shi, and Charlotte M Deane. Sabdab: the structural antibody database.Nucleic acids research, 42(D1):D1140–D1146, 2014

    2014

  3. [12]

    Openmm 7: Rapid development of high performance algorithms for molecular dynamics.PLoS computational biology, 13(7):e1005659, 2017

    Peter Eastman, Jason Swails, John D Chodera, Robert T McGibbon, Yutong Zhao, Kyle A Beauchamp, Lee-Ping Wang, Andrew C Simmonett, Matthew P Harrigan, Chaya D Stern, et al. Openmm 7: Rapid development of high performance algorithms for molecular dynamics.PLoS computational biol...

    2017

  4. [13]

    Pre-training antibody language models for antigen-specific computational antibody design

    Kaiyuan Gao, Lijun Wu, Jinhua Zhu, Tianbo Peng, Yingce Xia, Liang He, Shufang Xie, Tao Qin, Haiguang Liu, Kun He, et al. Pre-training antibody language models for antigen-specific computational antibody design. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Disco...

    2023

  5. [14]

    A review of deep learning methods for antibodies.Antibodies, 9(2):12, 2020

    Jordan Graves, Jacob Byerly, Eduardo Priego, Naren Makkapati, S Vince Parish, Brenda Medellin, and Monica Berrondo. A review of deep learning methods for antibodies.Antibodies, 9(2):12, 2020. 10

    2020

  6. [15]

    Measuring antibody–antigen binding kinetics using surface plasmon resonance.Antibody Engineering: Methods and Protocols, Second Edition, pages 411–442, 2012

    Stephen Hearty, Paul Leonard, and Richard O’Kennedy. Measuring antibody–antigen binding kinetics using surface plasmon resonance.Antibody Engineering: Methods and Protocols, Second Edition, pages 411–442, 2012

    2012

  7. [16]

    Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    2020

  8. [17]

    Argmax flows and multinomial diffusion: Learning categorical distributions.Advances in Neural Information Processing Systems, 34:12454–12465, 2021

    Emiel Hoogeboom, Didrik Nielsen, Priyank Jaini, Patrick Forré, and Max Welling. Argmax flows and multinomial diffusion: Learning categorical distributions.Advances in Neural Information Processing Systems, 34:12454–12465, 2021

    2021

  9. [18]

    Antibody-antigen docking and design via hierarchical structure refinement

    Wengong Jin, Regina Barzilay, and Tommi Jaakkola. Antibody-antigen docking and design via hierarchical structure refinement. InInternational Conference on Machine Learning, pages 10217–10227. PMLR, 2022

    2022

  10. [19]

    Exploiting sequence space: shuffling in vivo formed complementarity determining regions into a master framework.Gene, 215(2): 471–476, 1998

    P Jirholt, M Ohlin, CAK Borrebaeck, and E Söderlind. Exploiting sequence space: shuffling in vivo formed complementarity determining regions into a master framework.Gene, 215(2): 471–476, 1998

    1998

  11. [20]

    Structure based design of non-natural peptidic macrocyclic mcl-1 inhibitors.ACS medicinal chemistry letters, 8(2):239–244, 2017

    Jeffrey W Johannes, Stephanie Bates, Carl Beigie, Matthew A Belmonte, John Breen, Shenggen Cao, Paolo A Centrella, Matthew A Clark, John W Cuozzo, Christoph E Dumelin, et al. Structure based design of non-natural peptidic macrocyclic mcl-1 inhibitors.ACS medicinal chemistry le...

    2017

  12. [21]

    Replacing the complementarity-determining regions in a human antibody with those from a mouse.Nature, 321(6069):522–525, 1986

    Peter T Jones, Paul H Dear, Jefferson Foote, Michael S Neuberger, and Greg Winter. Replacing the complementarity-determining regions in a human antibody with those from a mouse.Nature, 321(6069):522–525, 1986

    1986

  13. [22]

    Ant colony sampling with gflownets for combinatorial optimization.arXiv preprint arXiv:2403.07041, 2024

    Minsu Kim, Sanghyeok Choi, Hyeonah Kim, Jiwoo Son, Jinkyoo Park, and Yoshua Ben- gio. Ant colony sampling with gflownets for combinatorial optimization.arXiv preprint arXiv:2403.07041, 2024

    2024 arXiv

  14. [23]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  15. [24]

    Conditional antibody design as 3d equivariant graph translation.arXiv preprint arXiv:2208.06073, 2022

    Xiangzhe Kong, Wenbing Huang, and Yang Liu. Conditional antibody design as 3d equivariant graph translation.arXiv preprint arXiv:2208.06073, 2022

    2022 arXiv

  16. [25]

    End-to-end full-atom antibody design.arXiv preprint arXiv:2302.00203, 2023

    Xiangzhe Kong, Wenbing Huang, and Yang Liu. End-to-end full-atom antibody design.arXiv preprint arXiv:2302.00203, 2023

    2023 arXiv

  17. [26]

    Abdesign: An algorithm for combinatorial backbone design guided by natural conformations and sequences.Proteins: Structure, Function, and Bioinformatics, 83 (8):1385–1406, 2015

    Gideon D Lapidoth, Dror Baran, Gabriele M Pszolla, Christoffer Norn, Assaf Alon, Michael D Tyka, and Sarel J Fleishman. Abdesign: An algorithm for combinatorial backbone design guided by natural conformations and sequences.Proteins: Structure, Function, and Bioinformatics, 83 ...

    2015

  18. [27]

    Degiacomi, and Chris G

    Adam Leach, Sebastian M Schmon, Matteo T. Degiacomi, and Chris G. Willcocks. Denoising diffusion probabilistic models on SO(3) for rotational alignment. InICLR 2022 Workshop on Geometrical and Topological Representation Learning, 2022. URL https://openreview. net/forum?id=BY88eBbkpe5

    2022

  19. [28]

    Language models of protein sequences at the scale of evolution enable accurate structure prediction.BioRxiv, 2022: 500902, 2022

    Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, et al. Language models of protein sequences at the scale of evolution enable accurate structure prediction.BioRxiv, 2022: 500902, 2022

    2022

  20. [29]

    Development of therapeutic antibodies for the treatment of diseases.Journal of biomedical science, 27:1–30, 2020

    Ruei-Min Lu, Yu-Chyi Hwang, I-Ju Liu, Chi-Chiu Lee, Han-Zen Tsai, Hsin-Jung Li, and Han-Chung Wu. Development of therapeutic antibodies for the treatment of diseases.Journal of biomedical science, 27:1–30, 2020

    2020

  21. [30]

    Antigen- specific antibody design and optimization with diffusion-based generative models for protein structures.bioRxiv, 2022

    Shitong Luo, Yufeng Su, Xingang Peng, Sheng Wang, Jian Peng, and Jianzhu Ma. Antigen- specific antibody design and optimization with diffusion-based generative models for protein structures.bioRxiv, 2022. doi: 10.1101/2022.07.10.499510. URL https://www.biorxiv. org/content/ear...

    2022 doi

  22. [31]

    Trajectory balance: Improved credit assignment in gflownets.Advances in Neural Information Processing Systems, 35:5955–5967, 2022

    Nikolay Malkin, Moksh Jain, Emmanuel Bengio, Chen Sun, and Yoshua Bengio. Trajectory balance: Improved credit assignment in gflownets.Advances in Neural Information Processing Systems, 35:5955–5967, 2022

    2022

  23. [32]

    Abdif- fuser: full-atom generation of in-vitro functioning antibodies.Advances in Neural Information Processing Systems, 36, 2024

    Karolis Martinkus, Jan Ludwiczak, Wei-Ching Liang, Julien Lafrance-Vanasse, Isidro Hotzel, Arvind Rajpal, Yan Wu, Kyunghyun Cho, Richard Bonneau, Vladimir Gligorijevic, et al. Abdif- fuser: full-atom generation of in-vitro functioning antibodies.Advances in Neural Information ...

    2024

  24. [33]

    Applications of surface plasmon resonance and biolayer interferometry for virus–ligand binding.Viruses, 14 (4):717, 2022

    Shruthi Murali, Richard R Rustandi, Xiwei Zheng, Anne Payne, and Liang Shang. Applications of surface plasmon resonance and biolayer interferometry for virus–ligand binding.Viruses, 14 (4):717, 2022

    2022

  25. [34]

    Origins of specificity and affinity in antibody–protein interactions.Proceedings of the National Academy of Sciences, 111 (26):E2656–E2665, 2014

    Hung-Pin Peng, Kuo Hao Lee, Jhih-Wei Jian, and An-Suei Yang. Origins of specificity and affinity in antibody–protein interactions.Proceedings of the National Academy of Sciences, 111 (26):E2656–E2665, 2014

    2014

  26. [35]

    Generative diffusion models for antibody design, docking, and optimization.bioRxiv, pages 2023–09, 2023

    Zhangzhi Peng, Chenchen Han, Xiaohan Wang, Dapeng Li, and Fajie Yuan. Generative diffusion models for antibody design, docking, and optimization.bioRxiv, pages 2023–09, 2023

    2023

  27. [36]

    Antibody complementarity-determining regions (cdrs) can display differential antimicrobial, antiviral and antitumor activities.PLoS One, 3(6):e2371, 2008

    Luciano Polonelli, José Pontón, Natalia Elguezabal, María Dolores Moragues, Claudio Casoli, Elisabetta Pilotti, Paola Ronzi, Andrey S Dobroff, Elaine G Rodrigues, Maria A Juliano, et al. Antibody complementarity-determining regions (cdrs) can display differential antimicrobial...

    2008

  28. [37]

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

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems, 36, 2024

    2024

  29. [38]

    Multi-objective antibody design with constrained preference optimization

    Milong Ren, ZaiKai He, and Haicang Zhang. Multi-objective antibody design with constrained preference optimization. InThe Thirteenth International Conference on Learning Representa- tions, 2025

    2025

  30. [39]

    Grinaway, David L

    Andrea Rizzi, Peter B. Grinaway, David L. Parton, Michael R. Shirts, K. Wang, Peter Eastman, Michael Friedrichs, Vijay S. Pande, Kim Branson, David L. Mobley, and John D. Chodera. Yank: A gpu-accelerated platform for alchemical free energy calculations. https://getyank.org, 2020

    2020

  31. [40]

    Gautier Robin, Yoshiteru Sato, Dominique Desplancq, Natacha Rochel, Etienne Weiss, and Pierre Martineau. Restricted diversity of antigen binding residues of antibodies revealed by computational alanine scanning of 227 antibody–antigen complexes.Journal of Molecular Biology, 42...

    2014

  32. [41]

    Mmseqs2: sensitive protein sequence searching for the analysis of massive data sets.bioRxiv, 2017

    Martin Steinegger and Johannes Söding. Mmseqs2: sensitive protein sequence searching for the analysis of massive data sets.bioRxiv, 2017. doi: 10.1101/079681. URL https: //www.biorxiv.org/content/early/2017/06/07/079681

    2017 doi

  33. [42]

    MIT press Cambridge, 1998

    Richard S Sutton, Andrew G Barto, et al.Reinforcement learning: An introduction, volume 1. MIT press Cambridge, 1998

    1998

  34. [43]

    A survey of generative ai for de novo drug design: New frontiers in molecule and protein generation, 2024

    Xiangru Tang, Howard Dai, Elizabeth Knight, Fang Wu, Yunyang Li, Tianxiao Li, and Mark Gerstein. A survey of generative ai for de novo drug design: New frontiers in molecule and protein generation, 2024. URLhttps://arxiv.org/abs/2402.08703

    2024 arXiv

  35. [44]

    Design of humanized antibodies: from anti-tac to zenapax.Methods, 36(1):69–83, 2005

    Naoya Tsurushita, Paul R Hinton, and Shankar Kumar. Design of humanized antibodies: from anti-tac to zenapax.Methods, 36(1):69–83, 2005

    2005

  36. [45]

    Amortizing intractable inference in diffusion models for vision, language, and control.arXiv preprint arXiv:2405.20971, 2024

    Siddarth Venkatraman, Moksh Jain, Luca Scimeca, Minsu Kim, Marcin Sendera, Mohsin Hasan, Luke Rowe, Sarthak Mittal, Pablo Lemos, Emmanuel Bengio, et al. Amortizing intractable inference in diffusion models for vision, language, and control.arXiv preprint arXiv:2405.20971, 2024

    2024 arXiv

  37. [46]

    Prediction of protein– protein binding free energies.Protein Science, 21(3):396–404, 2012

    Thom Vreven, Howook Hwang, Brian G Pierce, and Zhiping Weng. Prediction of protein– protein binding free energies.Protein Science, 21(3):396–404, 2012. 12

    2012

  38. [47]

    Alignab: Pareto-optimal energy alignment for designing nature-like antibodies.arXiv preprint arXiv:2412.20984, 2024

    Yibo Wen, Chenwei Xu, Jerry Yao-Chieh Hu, and Han Liu. Alignab: Pareto-optimal energy alignment for designing nature-like antibodies.arXiv preprint arXiv:2412.20984, 2024

    2024

  39. [48]

    Dinghuai Zhang, Ricky T. Q. Chen, Cheng-Hao Liu, Aaron Courville, and Yoshua Bengio. Diffusion generative flow samplers: Improving learning signals through partial trajectory optimization, 2024. URLhttps://arxiv.org/abs/2310.02679

    2024 arXiv

  40. [49]

    Improving gflownets for text-to-image diffusion alignment.arXiv preprint arXiv:2406.00633, 2024

    Dinghuai Zhang, Yizhe Zhang, Jiatao Gu, Ruixiang Zhang, Josh Susskind, Navdeep Jaitly, and Shuangfei Zhai. Improving gflownets for text-to-image diffusion alignment.arXiv preprint arXiv:2406.00633, 2024

    2024 arXiv

  41. [50]

    Packdock: a diffusion based side chain packing model for flexible protein-ligand docking.bioRxiv, pages 2024–01, 2024

    Runze Zhang, Xinyu Jiang, Duanhua Cao, Jie Yu, Mingan Chen, Zhehuan Fan, Xiangtai Kong, Jiacheng Xiong, Zimei Zhang, Wei Zhang, et al. Packdock: a diffusion based side chain packing model for flexible protein-ligand docking.bioRxiv, pages 2024–01, 2024

    2024

  42. [51]

    Diffpack: A torsional diffusion model for autoregressive protein side-chain packing.Advances in Neural Information Processing Systems, 36:48150–48172, 2023

    Yangtian Zhang, Zuobai Zhang, Bozitao Zhong, Sanchit Misra, and Jian Tang. Diffpack: A torsional diffusion model for autoregressive protein side-chain packing.Advances in Neural Information Processing Systems, 36:48150–48172, 2023

    2023

  43. [52]

    Neoantigen: a new breakthrough in tumor immunotherapy.Frontiers in immunology, 12:672356, 2021

    Zheying Zhang, Manman Lu, Yu Qin, Wuji Gao, Li Tao, Wei Su, and Jiateng Zhong. Neoantigen: a new breakthrough in tumor immunotherapy.Frontiers in immunology, 12:672356, 2021

    2021

  44. [53]

    Antigen-specific antibody design via direct energy-based preference optimization.arXiv preprint arXiv:2403.16576, 2024

    Xiangxin Zhou, Dongyu Xue, Ruizhe Chen, Zaixiang Zheng, Liang Wang, and Quanquan Gu. Antigen-specific antibody design via direct energy-based preference optimization.arXiv preprint arXiv:2403.16576, 2024

    2024 arXiv

  45. [54]

    Antibody design using a score-based diffusion model guided by evolutionary, physical and geometric constraints

    Tian Zhu, Milong Ren, and Haicang Zhang. Antibody design using a score-based diffusion model guided by evolutionary, physical and geometric constraints. InForty-first International Conference on Machine Learning, 2024. 13 Technical Appendices and Supplementary Material A Exper...

    2024

  46. [55]

    The Trajectory Balance objective requires fully generating a CDR, which in our setup requires 100 forward passes with the neural network for each gradient update. 2) As thein vitroaffinity data for all training complexes is not available and for fair comparison with existing m...

Pith tools

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