Pith. sign in

REVIEW 4 major objections 4 minor 77 references

A multimodal antibody foundation model generates complete antigen-specific antibody designs conditioned on antigen geometry and epitope positions, and preference optimization against structural-confidence scores yields state-of-the-art in s

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 10:53 UTC pith:5AJAJUD6

load-bearing objection A well-engineered antibody design pipeline whose 'functional' claim is unsupported by its own evaluation loop, but the methodological core is worth a serious referee. the 4 major comments →

arxiv 2607.20057 v1 pith:5AJAJUD6 submitted 2026-07-22 q-bio.BM cs.LG

Antigen-specific Antibody Multi-modal Foundation Model for Functional Antibody Design

Xiaoliang Shi , Zichen Wang , Runze Ma , Zhongyue Zhang , Shuangjia Zheng This is my paper
classification q-bio.BM cs.LG
keywords antibody designantigen-specificmultimodal protein language modelcomplementarity-determining regionspreference optimizationepitope conditioningstructure prediction confidenceCDR co-design
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to show that a single multimodal antibody model can take an antigen structure, a specified epitope, and an antibody framework sequence, and jointly output the sequences of all six CDR loops plus the complete antibody structure. To do this it adapts a large general protein language model to antibody sequence–structure pairs, then injects antigen geometry and epitope information through a small cross-modal adapter. A final preference-optimization stage steers the model toward designs that score highly on structure-prediction confidence and on natural-antibody sequence plausibility. The paper reports that this pipeline, called AAMFM with Cal-DPO, outperforms previous antibody co-design methods on all four headline metrics (PLL, AF3 score, pTM, and ipTM) and on most CDR-level metrics. A sympathetic reader would care because it is a concrete route from an antigen's structure to plausible, antigen-specific antibody candidates without experimental screening.

Core claim

AAMFM's central claim is that unified representation learning over antibody sequence, antibody structure, and antigen context—rather than sequence-only language modeling or graph-only co-design—is what makes antigen-specific functional design work. The model is built on a multimodal protein language model, post-trained on roughly 1.4 million paired antibody sequences and structures, then fine-tuned on experimental antibody–antigen complexes. The Antigen Geometric-Epitope-aware Adapter, only about 5.1 million parameters, fuses graph-extracted antigen geometry and epitope-interface embeddings into the model's latent space via residual adjustment. The Cal-DPO stage then optimizes against a pref

What carries the argument

The load-bearing mechanism is the combination of the Antigen Geometric-Epitope-aware Adapter and the Cal-DPO preference-alignment stage. The adapter is a lightweight residual module that conditions the base model on antigen geometry (via a graph neural network) and epitope positions (via an interface encoder), letting the model attend to the binding site during generation. Cal-DPO is a preference-optimization loss that, unlike plain DPO, also pulls the absolute log-likelihood ratios of preferred and dispreferred CDR sequences toward a fixed margin, so the model's confidence tracks the oracle signal instead of just ranking pairs. The oracle signal itself is the AF3 ranking/confidence score fr

Load-bearing premise

The central claim collapses if Protenix/AF3 confidence scores and ipTM are only weakly correlated with true antibody–antigen binding: the whole preference signal and the headline evaluation metrics are built on those structural-confidence scores, and the paper explicitly notes that no in vitro validation of the designed antibodies was performed.

What would settle it

Collect a panel of 50–100 AAMFM-CalDPO designs spanning a range of predicted AF3 scores for one or two antigens, synthesize and express them, and measure binding by surface plasmon resonance or yeast display; if the rank order of measured affinities does not track the predicted AF3 score and ipTM, the functional-design claim is refuted even if all in silico metrics replicate.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Given an antigen structure and epitope mask, AAMFM-CalDPO can produce full CDR sequences and antibody structure in one generation pass, removing the need for separate sequence and structure models.
  • The Cal-DPO stage intentionally trades some native-sequence recovery (AAR) for higher predicted binding confidence and plausibility, so users optimizing for binding can use the preference-aligned variant while users optimizing for human-likeness can use the SFT variant.
  • Ablations show that removing the antigen adapter causes the largest drop in functional metrics, and disabling epitope input degrades foldability and binding confidence—so antigen geometry and epitope conditioning, not just the language-model prior, carry the antigen-specific signal.
  • The released preference dataset of roughly 30,000 pairs with AF3 and PLL labels is a reusable resource for training or aligning other antibody design models.
  • For CDR-H3-only redesign, the same model beats dedicated CDR-H3 co-design baselines on predicted-binding metrics, suggesting the framework applies at both full-antibody and single-loop granularity.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If AF3 confidence scores and ipTM correlate strongly with real binding—which the paper assumes but does not test—then AAMFM's preference pipeline is a high-throughput in silico prefilter that could substantially shrink the sequence space needing wet-lab screening.
  • The adapter-plus-preference recipe is not antibody-specific in principle; the same conditioning on partner geometry and interface labels could be applied to nanobodies, peptide binders, or other protein–protein interaction design tasks.
  • Because the preference labels currently come entirely from structural-prediction confidence, a natural and testable extension is to rebuild the preference dataset with experimentally measured affinities for a small set of antigens and check whether Cal-DPO then improves real binding rates rather than only predicted ones.
  • The dual-threshold labeling rule (simultaneous margins on AF3 and PLL) could be tuned or replaced with other quality signals—solubility, developability, immunogenicity—to steer generation toward a different definition of 'functional'.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper presents AAMFM, a multimodal antibody language model built on ESM3. An antigen geometric/epitope-aware adapter injects antigen structure and epitope information, followed by two fine-tuning stages: SFT on antibody-antigen complexes (SAbDab) and Cal-DPO using a preference dataset whose labels are derived from Protenix (an AlphaFold3 reproduction) AF3 scores and AntiBERTy pseudo-log-likelihood. Experiments on full-antibody co-design and CDR-H3 co-design compare against several baselines, reporting PLL, AF3-score, pTM, ipTM, AAR, RMSD and PHR. The main claim is state-of-the-art functional antibody design.

Significance. If the claims were well supported, the paper would be a meaningful contribution to computational antibody design: the multi-stage adaptation from a general protein foundation model, the lightweight antigen adapter, and the construction of a ~30k preference-pair dataset are all useful elements, and the code is promised to be public. However, the headline claim rests on an evaluation that is substantially circular: the training objective in §3.4 is the same family of metrics (Protenix AF3 score, AntiBERTy PLL) used as the primary evaluation axes in §4.1. Baselines were not trained against this oracle. Furthermore, no independent binding evidence is provided. Thus the current support for 'functional antibody design' is weak, although the methodological pieces could be salvaged with a major revision that adds independent validation or substantially reframes the claims.

major comments (4)
  1. [§3.4, §4.1, Table 1] The preference labels in §3.4 are defined by Protenix AF3 score and AntiBERTy PLL (dual thresholds 0.2 and 0.1). Table 1 then evaluates the model on exactly these metrics (PLL, AF3-score, pTM, ipTM). Since AAMFM-CalDPO is explicitly optimized to increase these scores, while the baselines are not, the reported gains largely measure reward optimization rather than functional superiority. This circularity is the central load-bearing issue. The authors should evaluate on an independent metric not used in training (e.g., experimentally measured binding affinities, or a held-out oracle) or clearly limit the claims to 'improving in-silico confidence scores'.
  2. [Abstract / Conclusion / §3.4] The abstract and §4.1 claim state-of-the-art 'functional antibody design,' but the Conclusion admits that no in vitro validation was performed. AF3-score and ipTM are structure-prediction confidence metrics, not binding measurements. The paper itself notes the risk of reward hacking in §3.4, and the cited correlations (refs [35,21,62]) do not establish that maximizing these confidence scores yields functional antibodies. Without an independent binding assay, the term 'functional' is an overclaim. This is not merely a presentation issue; it undermines the central contribution.
  3. [Appendix A.3, §4.1] The DPO preference dataset is generated from 'all PDB structures in the RAbD dataset' (Appendix A.3), while evaluation is reported on SAbDab (Tables 1–3). RAbD is derived from SAbDab, so it is unclear whether the complexes used for preference-data construction overlap with the evaluation set. If such overlap exists, the evaluation is further confounded by data leakage. The authors should specify the exact split and show that both SFT and DPO training data are disjoint from the test set.
  4. [Appendix A.1, Tables 1–3] Two evaluation-fairness issues: (a) AbX is evaluated with provided checkpoints rather than being retrained on the same dataset as the other baselines (DiffAb, dyMEAN, GeoAB are retrained), which could bias the comparison. (b) No error bars or significance tests are reported. With only 10 samples per antigen and no variance information, small differences such as PLL -0.87 vs -0.91 in Table 1 may be within noise. Confidence intervals or paired tests should be provided to support the claimed superiority.
minor comments (4)
  1. [Figure 1] Figure 1 is visually overloaded and the chain labels (e.g., 'H1FR1L2FR2L1FR1 L3FR3H3FR3H2FR2FR4FR4') are garbled and hard to read. A cleaner schematic would help the reader follow the pipeline.
  2. [Appendix D.2, Eq. (11)] Equation (11) writes the DPO loss as -E[log σ(β log πθ(yw)/πref(yw)) - log σ(β log πθ(yl)/πref(yl))], which is not the standard DPO loss. The main-text Eq. (3) is correct, so this is likely a typographical error; please fix it.
  3. [§3.4 'Preference Definition'] The dual thresholds (0.2 in AF3 score and 0.1 in PLL) are introduced without justification or sensitivity analysis. Please provide a rationale or a small study showing the results are robust to these choices.
  4. [Table 4 caption] The caption 'Ablation on CDRs design' is vague. Specify that this is the full six-CDR co-design setting consistent with §4.1, and clarify that the 'w/o DPO (SFT)' row is the same as AAMFM-SFT.

Circularity Check

2 steps flagged

Cal-DPO is trained on the same Protenix/AntiBERTy scores that later serve as the headline evaluation; the 'functional' claim is partly a report of the model optimizing its own oracle.

specific steps
  1. fitted input called prediction [§3.4 'Preference Definition' vs §4.1 'Metrics'/'Results']
    "Each sequence is then scored by Protenix [10], a reproduction of AlphaFold3, to obtain a structure-based estimate of antigen-binding confidence at the designed CDR–antigen interface, which we use as the primary oracle for antibody functionality. / the AF3 score, derived from predicting the complex of the designed antibody and antigen, which comprehensively reflects the design’s foldability and its antigen-binding functionality"

    The Cal-DPO preference labels are constructed from the same Protenix AF3 score that later serves as the headline functional metric. Since the AF3 'ranking score' is a composite involving pTM and ipTM, the claimed gains on AF3-score, pTM, and ipTM are all aligned with the training reward. Reporting these gains after Cal-DPO is therefore not an independent confirmation of functional quality; it is partly a measure of how well the model optimized its own oracle.

  2. fitted input called prediction [§3.4 'Preference Definition' vs §4.1 'Metrics' and Tables 1, 3, 4]
    "To counteract this, we introduce pseudo-log-likelihood (PLL) under the antibody language model AntiBERTy [51] as a sequence-level plausibility constraint / the pseudo-log-likelihood (Pll) calculated by AntiBERTy to gauge sequence plausibility"

    PLL under AntiBERTy is one of the two preference-label criteria in the dual-threshold rule, and the same 'Pll' score is used as the main sequence-plausibility evaluation metric. Cal-DPO explicitly increases the probability of sequences preferred by higher PLL, so the reported Pll improvements in Tables 1, 3, and 4 are expected from the training signal rather than independent evidence of plausibility.

full rationale

The architectural derivation is not circular: ESM3 pretraining, OAS domain adaptation, SAbDab SFT, and the antigen adapter are independent of the evaluation metrics. The circularity enters at the alignment/evaluation stage. In §3.4, the preference dataset is built by labeling candidates with Protenix AF3 score and AntiBERTy PLL; in §4.1 the very same scores — plus pTM and ipTM, which the AF3 ranking score composites — are the headline evaluation axes, and AAMFM-CalDPO is declared best on exactly these axes. That is a fitted-input-called-prediction pattern: the evaluator was used to construct the training labels. The paper also cites [35], which includes two of the present authors, to support the load-bearing premise that AF3 confidence correlates with binding quality; however, external refs [21,62] are also cited, so I do not treat this self-citation as independently load-bearing circularity. The Conclusion explicitly admits the designed antibodies 'have not yet been fully validated through in vitro experiments,' so no external binding measurement supports the functional claim. The non-reward metrics (AAR, RMSD, PHR) show mixed, non-uniform gains, strengthening the view that the central 'state-of-the-art functional design' claim rests substantially on the training-aligned oracle. Overall this is partial circularity — the model is not merely renaming its inputs, but the key functional claim is partly a report of optimizing its own evaluator — hence score 6.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 0 invented entities

The central claim rests on treating in-silico confidence scores (Protenix AF3-score, ipTM) and language-model plausibility (AntiBERTy PLL) as proxies for real antibody functionality. These proxies are themselves used as training rewards, creating the circularity documented above. All other assumptions are standard domain assumptions for computational antibody design: pretrained model transfer, predicted structure quality, and known epitope input.

free parameters (4)
  • AF3-score dual-threshold margin = 0.2
    Hand-chosen threshold in §3.4: a candidate is preferred only if it exceeds the rejected sequence by at least 0.2 in AF3 score. This determines the entire preference dataset.
  • AntiBERTy PLL dual-threshold margin = 0.1
    Hand-chosen threshold in §3.4: candidate must also exceed rejected sequence by at least 0.1 in PLL. This shapes the preference signal.
  • Cal-DPO calibration loss weight λ = 1
    Set in §3.4: 'with a weighting factor λ = 1 for L_Cal'. Chosen by hand, not tuned or derived.
  • DPO temperature β
    Appears in Eq. 3 and defines the target margin M = 1/(2β), but no numerical value is reported anywhere in the paper, leaving the preference sharpness unspecified.
axioms (6)
  • domain assumption Protenix/AlphaFold3 confidence scores (AF3-score, ipTM, ranking score) are valid proxies for antibody-antigen binding functionality.
    Invoked in §3.4 as 'the primary oracle for antibody functionality' and in §4.1 as the core evaluation metric. Cited literature suggests correlation with interface quality, but the paper does not validate this proxy against experimental binding for its own generated sequences.
  • domain assumption AntiBERTy pseudo-log-likelihood is a valid measure of antibody sequence plausibility/naturalness.
    Used both as a preference label component (§3.4) and as an evaluation metric (§4.1). No independent calibration of what PLL differences mean for real antibody function is provided.
  • domain assumption ESM3's pretrained multimodal protein representations transfer meaningfully to antibody sequence-structure co-design after domain adaptation.
    The whole method builds on ESM3-open as the base model (§3.2, §4.1). The paper assumes the base model's learned prior is a beneficial starting point for antibody CDR design.
  • domain assumption IgFold and AbodyBuilder2 predicted structures for OAS paired sequences are accurate enough for antibody domain adaptation.
    Used in §3.2 to create 1.4M paired sequence-structure training examples. Errors in these predicted structures could propagate into the adapted model.
  • domain assumption Epitope positions on the target antigen are known and provided as input at inference.
    Task formulation in §3.1 includes epitope position I as given; the adapter encodes it as binary masks. In real design settings epitope information may be unavailable or uncertain, which is not addressed.
  • domain assumption Protenix is a sufficiently faithful reproduction of AlphaFold3 for scoring designed antibody-antigen complexes.
    The preference oracle and evaluation rely on Protenix rather than AlphaFold3 itself (§3.4, §A.2). Any systematic discrepancy between Protenix and AlphaFold3 confidence scores affects both training and evaluation.

pith-pipeline@v1.3.0-alltime-deepseek · 16701 in / 12704 out tokens · 114741 ms · 2026-08-01T10:53:50.682573+00:00 · methodology

0 comments
read the original abstract

Antibodies are essential proteins that play a central role in immune recognition by binding specific antigen molecules. Although recent protein language models have enabled progress in single-chain protein modeling and generation, they often fall short in antigen-specific antibody design, where effective modeling requires explicit pairing between antibody and antigen, particularly at the epitope level. To address these limitations, we introduce AAMFM, an Antigen-specific Antibody Multimodal Foundation Model that learns unified representations of antibody sequences and structures conditioned on antigen context. AAMFM incorporates rich antigen information including geometric interfaces and epitope annotations via a cross-modal adapter, enabling joint modeling of antibody-antigen interactions in a shared latent space. To further guide the model toward functional relevance, we fine-tune AAMFM using Calibrated Direct Preference Optimization (Cal-DPO), leveraging preference signals extracted from a strong structural prior to align learning with binding-specific objectives. Extensive experiments demonstrate that AAMFM achieves state-of-the-art performance in functional antibody design, revealing its potential for antigen-specific antibody engineering. Our code is available at https://github.com/XL-S224/AAMFM.

Figures

Figures reproduced from arXiv: 2607.20057 by Runze Ma, Shuangjia Zheng, Xiaoliang Shi, Zhongyue Zhang, Zichen Wang.

Figure 1
Figure 1. Figure 1: Hierarchical training and inference pipeline of the proposed model, including pre-training, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the dataset curation pro [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of the AF3 predicted structure of [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Training dynamic comparison between DPO and Cal-DPO [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

77 extracted references · 16 linked inside Pith

  1. [1]

    Immunebuilder: Deep-learning models for predicting the structures of immune proteins.Communications Biology, 6(1):575, 2023

    Brennan Abanades, Wing Ki Wong, Fergus Boyles, Guy Georges, Alexander Bujotzek, and Charlotte M Deane. Immunebuilder: Deep-learning models for predicting the structures of immune proteins.Communications Biology, 6(1):575, 2023

    2023

  2. [2]

    Accurate structure prediction of biomolecular interactions with alphafold 3.Nature, 630(8016):493–500, 2024

    Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J Ballard, Joshua Bambrick, et al. Accurate structure prediction of biomolecular interactions with alphafold 3.Nature, 630(8016):493–500, 2024

    2024

  3. [3]

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

    2018

  4. [4]

    A compact vocabulary of paratope-epitope interactions enables predictability of antibody-antigen binding.Cell Reports, 34(11), 2021

    Rahmad Akbar, Philippe A Robert, Milena Pavlovi´c, Jeliazko R Jeliazkov, Igor Snapkov, Andrei Slabodkin, Cédric R Weber, Lonneke Scheffer, Enkelejda Miho, Ingrid Hobæk Haff, et al. A compact vocabulary of paratope-epitope interactions enables predictability of antibody-antigen binding.Cell Reports, 34(11), 2021

    2021

  5. [5]

    Standard conformations for the canonical structures of immunoglobulins.Journal of molecular biology, 273(4):927–948, 1997

    Bissan Al-Lazikani, Arthur M Lesk, and Cyrus Chothia. Standard conformations for the canonical structures of immunoglobulins.Journal of molecular biology, 273(4):927–948, 1997

    1997

  6. [6]

    The rosetta all-atom energy function for macromolecular modeling and design

    Rebecca F Alford, Andrew Leaver-Fay, Jeliazko R Jeliazkov, Matthew J O’Meara, Frank P DiMaio, Hahnbeom Park, Maxim V Shapovalov, P Douglas Renfrew, Vikram K Mulligan, Kalli Kappel, et al. The rosetta all-atom energy function for macromolecular modeling and design. Journal of chemical theory and computation, 13(6):3031–3048, 2017

    2017

  7. [7]

    Direct preference optimization with an offset

    Afra Amini, Tim Vieira, and Ryan Cotterell. Direct preference optimization with an offset. arXiv preprint arXiv:2402.10571, 2024

    Pith/arXiv arXiv 2024

  8. [8]

    A general theoretical paradigm to understand learning from human preferences

    Mohammad Gheshlaghi Azar, Zhaohan Daniel Guo, Bilal Piot, Remi Munos, Mark Rowland, Michal Valko, and Daniele Calandriello. A general theoretical paradigm to understand learning from human preferences. InInternational Conference on Artificial Intelligence and Statistics, pages 4447–4455. PMLR, 2024

    2024

  9. [9]

    A surface expression vector for antibody screening.Gene, 104(2):147–153, 1991

    Frank Breitling, Stefan Dübel, Thomas Seehaus, Iris Klewinghaus, and Melvyn Little. A surface expression vector for antibody screening.Gene, 104(2):147–153, 1991

    1991

  10. [10]

    Protenix-advancing structure prediction through a comprehensive alphafold3 reproduction.bioRxiv, pages 2025–01, 2025

    AML AI4Science Team ByteDance, Xinshi Chen, Yuxuan Zhang, Chan Lu, Wenzhi Ma, Jiaqi Guan, Chengyue Gong, Jincai Yang, Hanyu Zhang, Ke Zhang, et al. Protenix-advancing structure prediction through a comprehensive alphafold3 reproduction.bioRxiv, pages 2025–01, 2025

    2025

  11. [11]

    Nanobody–antigen interaction prediction with ensemble deep learning and prompt-based protein language models.Nature Machine Intelligence, 6(12):1594–1604, 2024

    Juntao Deng, Miao Gu, Pengyan Zhang, Mingyu Dong, Tao Liu, Yabin Zhang, and Min Liu. Nanobody–antigen interaction prediction with ensemble deep learning and prompt-based protein language models.Nature Machine Intelligence, 6(12):1594–1604, 2024

    2024

  12. [12]

    Inverse folding for antibody sequence design using deep learning.arXiv preprint arXiv:2310.19513, 2023

    Frédéric A Dreyer, Daniel Cutting, Constantin Schneider, Henry Kenlay, and Charlotte M Deane. Inverse folding for antibody sequence design using deep learning.arXiv preprint arXiv:2310.19513, 2023

    Pith/arXiv arXiv 2023

  13. [13]

    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

  14. [14]

    Prottrans: Toward understanding the language of life through self-supervised learning.IEEE transactions on pattern analysis and machine intelligence, 44(10):7112–7127, 2021

    Ahmed Elnaggar, Michael Heinzinger, Christian Dallago, Ghalia Rehawi, Yu Wang, Llion Jones, Tom Gibbs, Tamas Feher, Christoph Angerer, Martin Steinegger, et al. Prottrans: Toward understanding the language of life through self-supervised learning.IEEE transactions on pattern analysis and machine intelligence, 44(10):7112–7127, 2021

    2021

  15. [15]

    Kto: Model alignment as prospect theoretic optimization.arXiv preprint arXiv:2402.01306, 2024

    Kawin Ethayarajh, Winnie Xu, Niklas Muennighoff, Dan Jurafsky, and Douwe Kiela. Kto: Model alignment as prospect theoretic optimization.arXiv preprint arXiv:2402.01306, 2024. 10

    Pith/arXiv arXiv 2024

  16. [16]

    Stefan Ewert, Annemarie Honegger, and Andreas Plückthun. Stability improvement of anti- bodies for extracellular and intracellular applications: Cdr grafting to stable frameworks and structure-based framework engineering.Methods, 34(2):184–199, 2004

    2004

  17. [17]

    Protgpt2 is a deep unsupervised language model for protein design.Nature communications, 13(1):4348, 2022

    Noelia Ferruz, Steffen Schmidt, and Birte Höcker. Protgpt2 is a deep unsupervised language model for protein design.Nature communications, 13(1):4348, 2022

    2022

  18. [18]

    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 Discovery and Data Mining, pages 506–517, 2023

    2023

  19. [19]

    Simulating 500 million years of evolution with a language model.Science, page eads0018, 2025

    Thomas Hayes, Roshan Rao, Halil Akin, Nicholas J Sofroniew, Deniz Oktay, Zeming Lin, Robert Verkuil, Vincent Q Tran, Jonathan Deaton, Marius Wiggert, et al. Simulating 500 million years of evolution with a language model.Science, page eads0018, 2025

    2025

  20. [20]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    2016

  21. [21]

    What does alphafold3 learn about antibody and nanobody docking, and what remains unsolved? InMAbs, volume 17, page 2545601

    Fatima N Hitawala and Jeffrey J Gray. What does alphafold3 learn about antibody and nanobody docking, and what remains unsolved? InMAbs, volume 17, page 2545601. Taylor & Francis, 2025

    2025

  22. [22]

    Antifold: Improved antibody structure-based design using inverse folding.arXiv preprint arXiv:2405.03370, 2024

    Magnus Haraldson Høie, Alissa Hummer, Tobias H Olsen, Broncio Aguilar-Sanjuan, Morten Nielsen, and Charlotte M Deane. Antifold: Improved antibody structure-based design using inverse folding.arXiv preprint arXiv:2405.03370, 2024

    Pith/arXiv arXiv 2024

  23. [23]

    Selecting and screening recombinant antibody libraries.Nature biotechnology, 23(9):1105–1116, 2005

    Hennie R Hoogenboom. Selecting and screening recombinant antibody libraries.Nature biotechnology, 23(9):1105–1116, 2005

    2005

  24. [24]

    Parameter-efficient transfer learning for nlp

    Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin De Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. Parameter-efficient transfer learning for nlp. InInternational conference on machine learning, pages 2790–2799. PMLR, 2019

    2019

  25. [25]

    Iterative refinement graph neural network for antibody sequence-structure co-design

    Wengong Jin, Jeremy Wohlwend, Regina Barzilay, and Tommi S Jaakkola. Iterative refinement graph neural network for antibody sequence-structure co-design. InInternational Conference on Learning Representations, 2022

    2022

  26. [26]

    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

  27. [27]

    Conditional antibody design as 3d equivariant graph translation

    Xiangzhe Kong, Wenbing Huang, and Yang Liu. Conditional antibody design as 3d equivariant graph translation. InThe Eleventh International Conference on Learning Representations, 2023

    2023

  28. [28]

    End-to-end full-atom antibody design

    Xiangzhe Kong, Wenbing Huang, and Yang Liu. End-to-end full-atom antibody design. In Proceedings of the 40th International Conference on Machine Learning, pages 17409–17429, 2023

    2023

  29. [29]

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

    2015

  30. [30]

    Geoab: Towards realistic antibody design and reliable affinity maturation.bioRxiv, pages 2024–05, 2024

    Haitao Lin, Lirong Wu, Yufei Huang, Yunfan Liu, Odin Zhang, Yuanqing Zhou, Rui Sun, and Stan Z Li. Geoab: Towards realistic antibody design and reliable affinity maturation.bioRxiv, pages 2024–05, 2024

    2024

  31. [31]

    Evolutionary-scale prediction of atomic-level protein structure with a language model.Science, 379(6637):1123–1130, 2023

    Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Nikita Smetanin, Robert Verkuil, Ori Kabeli, Yaniv Shmueli, et al. Evolutionary-scale prediction of atomic-level protein structure with a language model.Science, 379(6637):1123–1130, 2023. 11

    2023

  32. [32]

    Tis-dpo: Token-level importance sampling for direct preference optimization with estimated weights.arXiv preprint arXiv:2410.04350, 2024

    Aiwei Liu, Haoping Bai, Zhiyun Lu, Yanchao Sun, Xiang Kong, Simon Wang, Jiulong Shan, Al- bin Madappally Jose, Xiaojiang Liu, Lijie Wen, et al. Tis-dpo: Token-level importance sampling for direct preference optimization with estimated weights.arXiv preprint arXiv:2410.04350, 2024

    Pith/arXiv arXiv 2024

  33. [33]

    Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

    2023

  34. [34]

    Llava-plus: Learning to use tools for creating multimodal agents

    Shilong Liu, Hao Cheng, Haotian Liu, Hao Zhang, Feng Li, Tianhe Ren, Xueyan Zou, Jianwei Yang, Hang Su, Jun Zhu, et al. Llava-plus: Learning to use tools for creating multimodal agents. InEuropean Conference on Computer Vision, pages 126–142. Springer, 2024

    2024

  35. [35]

    Alphafold3, a secret sauce for predicting mutational effects on protein-protein interactions.bioRxiv, pages 2024–05, 2024

    Wei Lu, Jixian Zhang, Jiahua Rao, Zhongyue Zhang, and Shuangjia Zheng. Alphafold3, a secret sauce for predicting mutational effects on protein-protein interactions.bioRxiv, pages 2024–05, 2024

    2024

  36. [36]

    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.Advances in Neural Information Processing Systems, 35:9754–9767, 2022

    2022

  37. [37]

    Rotamer density estimator is an unsupervised learner of the effect of mutations on protein-protein interaction

    Shitong Luo, Yufeng Su, Zuofan Wu, Chenpeng Su, Jian Peng, and Jianzhu Ma. Rotamer density estimator is an unsupervised learner of the effect of mutations on protein-protein interaction. bioRxiv, pages 2023–02, 2023

    2023

  38. [38]

    Abdif- fuser: full-atom generation of in-vitro functioning antibodies.Advances in Neural Information Processing Systems, 36:40729–40759, 2023

    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 Processing Systems, 36:40729–40759, 2023

    2023

  39. [39]

    Language models enable zero-shot prediction of the effects of mutations on protein function.Advances in neural information processing systems, 34:29287–29303, 2021

    Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu, and Alex Rives. Language models enable zero-shot prediction of the effects of mutations on protein function.Advances in neural information processing systems, 34:29287–29303, 2021

    2021

  40. [40]

    Simpo: Simple preference optimization with a reference-free reward.Advances in Neural Information Processing Systems, 37:124198–124235, 2024

    Yu Meng, Mengzhou Xia, and Danqi Chen. Simpo: Simple preference optimization with a reference-free reward.Advances in Neural Information Processing Systems, 37:124198–124235, 2024

    2024

  41. [41]

    Progen2: exploring the boundaries of protein language models.Cell systems, 14(11):968–978, 2023

    Erik Nijkamp, Jeffrey A Ruffolo, Eli N Weinstein, Nikhil Naik, and Ali Madani. Progen2: exploring the boundaries of protein language models.Cell systems, 14(11):968–978, 2023

    2023

  42. [42]

    Ablang: an antibody language model for completing antibody sequences.Bioinformatics Advances, 2(1):vbac046, 2022

    Tobias H Olsen, Iain H Moal, and Charlotte M Deane. Ablang: an antibody language model for completing antibody sequences.Bioinformatics Advances, 2(1):vbac046, 2022

    2022

  43. [43]

    Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    2022

  44. [44]

    Smaug: Fixing failure modes of preference optimisation with dpo-positive.arXiv preprint arXiv:2402.13228, 2024

    Arka Pal, Deep Karkhanis, Samuel Dooley, Manley Roberts, Siddartha Naidu, and Colin White. Smaug: Fixing failure modes of preference optimisation with dpo-positive.arXiv preprint arXiv:2402.13228, 2024

    Pith/arXiv arXiv 2024

  45. [45]

    Iterative reasoning preference optimization.Advances in Neural Information Processing Systems, 37:116617–116637, 2024

    Richard Yuanzhe Pang, Weizhe Yuan, He He, Kyunghyun Cho, Sainbayar Sukhbaatar, and Jason Weston. Iterative reasoning preference optimization.Advances in Neural Information Processing Systems, 37:116617–116637, 2024

    2024

  46. [46]

    Antibody engineering.Current Opinion in Structural Biology, 2(4):593–596, 1992

    Leonard G Presta. Antibody engineering.Current Opinion in Structural Biology, 2(4):593–596, 1992

    1992

  47. [47]

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

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems, 36:53728–53741, 2023. 12

    2023

  48. [48]

    Msa transformer

    Roshan M Rao, Jason Liu, Robert Verkuil, Joshua Meier, John Canny, Pieter Abbeel, Tom Sercu, and Alexander Rives. Msa transformer. InInternational Conference on Machine Learning, pages 8844–8856. PMLR, 2021

    2021

  49. [49]

    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

  50. [50]

    Fast, accurate antibody structure prediction from deep learning on massive set of natural antibodies.Nature communications, 14(1):2389, 2023

    Jeffrey A Ruffolo, Lee-Shin Chu, Sai Pooja Mahajan, and Jeffrey J Gray. Fast, accurate antibody structure prediction from deep learning on massive set of natural antibodies.Nature communications, 14(1):2389, 2023

    2023

  51. [51]

    Deciphering antibody affinity maturation with language models and weakly supervised learning.arXiv preprint arXiv:2112.07782, 2021

    Jeffrey A Ruffolo, Jeffrey J Gray, and Jeremias Sulam. Deciphering antibody affinity maturation with language models and weakly supervised learning.arXiv preprint arXiv:2112.07782, 2021

    Pith/arXiv arXiv 2021

  52. [52]

    Design of highly functional genome editors by modeling the universe of crispr-cas sequences.BioRxiv, pages 2024–04, 2024

    Jeffrey A Ruffolo, Stephen Nayfach, Joseph Gallagher, Aadyot Bhatnagar, Joel Beazer, Riffat Hussain, Jordan Russ, Jennifer Yip, Emily Hill, Martin Pacesa, et al. Design of highly functional genome editors by modeling the universe of crispr-cas sequences.BioRxiv, pages 2024–04, 2024

    2024

  53. [53]

    Aipo: Improving training objective for iterative preference optimization.arXiv preprint arXiv:2409.08845, 2024

    Yaojie Shen, Xinyao Wang, Yulei Niu, Ying Zhou, Lexin Tang, Libo Zhang, Fan Chen, and Longyin Wen. Aipo: Improving training objective for iterative preference optimization.arXiv preprint arXiv:2409.08845, 2024

    Pith/arXiv arXiv 2024

  54. [54]

    Moviellm: Enhancing long video understanding with ai-generated movies.arXiv preprint arXiv:2403.01422, 2024

    Zhende Song, Chenchen Wang, Jiamu Sheng, Chi Zhang, Gang Yu, Jiayuan Fan, and Tao Chen. Moviellm: Enhancing long video understanding with ai-generated movies.arXiv preprint arXiv:2403.01422, 2024

    Pith/arXiv arXiv 2024

  55. [55]

    Saprot: Protein language modeling with structure-aware vocabulary.bioRxiv, pages 2023–10, 2023

    Jin Su, Chenchen Han, Yuyang Zhou, Junjie Shan, Xibin Zhou, and Fajie Yuan. Saprot: Protein language modeling with structure-aware vocabulary.bioRxiv, pages 2023–10, 2023

    2023

  56. [56]

    Video understanding with large language models: A survey.IEEE Transactions on Circuits and Systems for Video Technology, 2025

    Yunlong Tang, Jing Bi, Siting Xu, Luchuan Song, Susan Liang, Teng Wang, Daoan Zhang, Jie An, Jingyang Lin, Rongyi Zhu, et al. Video understanding with large language models: A survey.IEEE Transactions on Circuits and Systems for Video Technology, 2025

    2025

  57. [57]

    Neural discrete representation learning.Advances in neural information processing systems, 30, 2017

    Aaron Van Den Oord, Oriol Vinyals, et al. Neural discrete representation learning.Advances in neural information processing systems, 30, 2017

    2017

  58. [58]

    On pre-training language model for antibody

    Danqing Wang, YE Fei, and Hao Zhou. On pre-training language model for antibody. InThe eleventh international conference on learning representations, 2023

    2023

  59. [59]

    Iggm: A generative model for functional antibody and nanobody design.bioRxiv, pages 2024–09, 2024

    Rubo Wang, Fandi Wu, Xingyu Gao, Jiaxiang Wu, Peilin Zhao, and Jianhua Yao. Iggm: A generative model for functional antibody and nanobody design.bioRxiv, pages 2024–09, 2024

    2024

  60. [60]

    Dplm-2: A multimodal diffusion protein language model

    Xinyou Wang, Zaixiang Zheng, YE Fei, Dongyu Xue, Shujian Huang, and Quanquan Gu. Dplm-2: A multimodal diffusion protein language model. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  61. [61]

    Retrieval augmented diffusion model for structure-informed antibody design and optimization.The Thirteenth International Conference on Learning Representations, 2025

    Zichen Wang, Yaokun Ji, Jianing Tian, and Shuangjia Zheng. Retrieval augmented diffusion model for structure-informed antibody design and optimization.The Thirteenth International Conference on Learning Representations, 2025

    2025

  62. [62]

    Benchmarking alphafold3’s protein-protein complex accuracy and machine learning prediction reliability for binding free energy changes upon mutation

    JunJie Wee and Guo-Wei Wei. Benchmarking alphafold3’s protein-protein complex accuracy and machine learning prediction reliability for binding free energy changes upon mutation. arXiv preprint arXiv:2406.03979, 2024

    Pith/arXiv arXiv 2024

  63. [63]

    Aligning protein generative models with experimental fitness via direct preference optimization.bioRxiv, pages 2024–05, 2024

    Talal Widatalla, Rafael Rafailov, and Brian Hie. Aligning protein generative models with experimental fitness via direct preference optimization.bioRxiv, pages 2024–05, 2024

    2024

  64. [64]

    Cal-dpo: Calibrated direct preference optimization for language model alignment.Advances in Neural Information Processing Systems, 37:114289–114320, 2024

    Teng Xiao, Yige Yuan, Huaisheng Zhu, Mingxiao Li, and Vasant Honavar. Cal-dpo: Calibrated direct preference optimization for language model alignment.Advances in Neural Information Processing Systems, 37:114289–114320, 2024. 13

    2024

  65. [65]

    A comprehensive survey of direct preference optimization: Datasets, theories, variants, and applications.arXiv preprint arXiv:2410.15595, 2024

    Wenyi Xiao, Zechuan Wang, Leilei Gan, Shuai Zhao, Wanggui He, Luu Anh Tuan, Long Chen, Hao Jiang, Zhou Zhao, and Fei Wu. A comprehensive survey of direct preference optimization: Datasets, theories, variants, and applications.arXiv preprint arXiv:2410.15595, 2024

    Pith/arXiv arXiv 2024

  66. [66]

    Aic mllm: Autonomous interactive correction mllm for robust robotic manipulation

    Chuyan Xiong, Chengyu Shen, Xiaoqi Li, Kaichen Zhou, Jeremy Liu, Ruiping Wang, and Hao Dong. Aic mllm: Autonomous interactive correction mllm for robust robotic manipulation. arXiv preprint arXiv:2406.11548, 2024

    Pith/arXiv arXiv 2024

  67. [67]

    Mlevlm: Improve multi-level progressive capabilities based on multimodal large language model for medical visual question answering

    Dexuan Xu, Yanyuan Chen, Jieyi Wang, Yue Huang, Hanpin Wang, Zhi Jin, Hongxing Wang, Weihua Yue, Jing He, Hang Li, et al. Mlevlm: Improve multi-level progressive capabilities based on multimodal large language model for medical visual question answering. InFindings of the Association for Computational Linguistics ACL 2024, pages 4977–4997, 2024

    2024

  68. [68]

    Diversity in the cdr3 region of vh is sufficient for most antibody specificities.Immunity, 13(1):37–45, 2000

    John L Xu and Mark M Davis. Diversity in the cdr3 region of vh is sufficient for most antibody specificities.Immunity, 13(1):37–45, 2000

    2000

  69. [69]

    Proteinbench: A holistic evaluation of protein foundation models.arXiv preprint arXiv:2409.06744, 2024

    Fei Ye, Zaixiang Zheng, Dongyu Xue, Yuning Shen, Lihao Wang, Yiming Ma, Yan Wang, Xinyou Wang, Xiangxin Zhou, and Quanquan Gu. Proteinbench: A holistic evaluation of protein foundation models.arXiv preprint arXiv:2409.06744, 2024

    Pith/arXiv arXiv 2024

  70. [70]

    S 2alm: Sequence-structure pre-trained large language model for comprehensive antibody representation learning.arXiv preprint arXiv:2411.15215, 2024

    Mingze Yin, Hanjing Zhou, Jialu Wu, Yiheng Zhu, Yuxuan Zhan, Zitai Kong, Hongxia Xu, Chang-Yu Hsieh, Jintai Chen, Tingjun Hou, et al. S 2alm: Sequence-structure pre-trained large language model for comprehensive antibody representation learning.arXiv preprint arXiv:2411.15215, 2024

    Pith/arXiv arXiv 2024

  71. [71]

    Protein representation learning by geometric structure pretraining

    Zuobai Zhang, Minghao Xu, Arian Rokkum Jamasb, Vijil Chenthamarakshan, Aurelie Lozano, Payel Das, and Jian Tang. Protein representation learning by geometric structure pretraining. In The Eleventh International Conference on Learning Representations, 2023

    2023

  72. [72]

    Sce2drivex: A generalized mllm framework for scene-to-drive learning.arXiv preprint arXiv:2502.14917, 2025

    Rui Zhao, Qirui Yuan, Jinyu Li, Haofeng Hu, Yun Li, Chengyuan Zheng, and Fei Gao. Sce2drivex: A generalized mllm framework for scene-to-drive learning.arXiv preprint arXiv:2502.14917, 2025

    Pith/arXiv arXiv 2025

  73. [73]

    Structure- informed language models are protein designers

    Zaixiang Zheng, Yifan Deng, Dongyu Xue, Yi Zhou, Fei Ye, and Quanquan Gu. Structure- informed language models are protein designers. InInternational conference on machine learning, pages 42317–42338. PMLR, 2023

    2023

  74. [74]

    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

    Pith/arXiv arXiv 2024

  75. [75]

    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. 14 A Experiment Details A.1 Baseline Details DiffAb[ 36] We used DiffAb from the official GitHub repository (https://github.com/luost26/di...

    arXiv 2024

  76. [76]

    Similarly, DPOP [44] and AIPO

    explicitly incorporates the negative log-likelihood (NLL) of preferred samples into the loss, thereby increasing the probability of generating preferred outputs. Similarly, DPOP [44] and AIPO

  77. [77]

    In contrast to these methods that only emphasize the preferred samples, Cal-DPO [64] simultaneously considers the absolute values of both preferred and dispreferred samples

    add extra loss terms based on the Bradley-Terry model to prevent the probability of preferred samples from decreasing during training. In contrast to these methods that only emphasize the preferred samples, Cal-DPO [64] simultaneously considers the absolute values of both preferred and dispreferred samples. It enables the model to learn implicit rewards i...