pith. sign in

arxiv: 2605.21485 · v1 · pith:7AHW3ZIInew · submitted 2026-05-20 · 💻 cs.LG

EvoStruct: Bridging Evolutionary and Structural Priors for Antibody CDR Design via Protein Language Model Adaptation

Mansoor Ahmed , Sujin Lee , Umar Khayaz , Murray Patterson This is my paper

Pith reviewed 2026-05-21 05:02 UTC · model grok-4.3

classification 💻 cs.LG
keywords antibody designCDR designprotein language modelgraph neural networksequence recoveryvocabulary collapsecross-attention adapterevolutionary priors
0
0 comments X

The pith

EvoStruct fuses a protein language model with structural GNN context through a cross-attention adapter to fix vocabulary collapse in antibody CDR design.

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

Current GNN methods for antibody complementarity-determining region design achieve high sequence recovery yet collapse to predicting only a narrow set of amino acids such as tyrosine and glycine. The paper traces this to the encoders learning distributions solely from limited structural examples and therefore losing substitution patterns present in evolutionary sequence databases. EvoStruct addresses the gap by connecting a frozen protein language model to an E(3)-equivariant GNN via a cross-attention adapter, then progressively unfreezing the language model layers under R-Drop consistency regularization. On the CHIMERA-Bench dataset the approach records the highest amino acid recovery, lowest perplexity, 2.3 times greater sequence diversity, and strongest correlation with ground-truth binding pairs. A sympathetic reader would care because greater functional diversity in designed CDRs directly expands the pool of viable therapeutic candidates that can be advanced to experimental testing.

Core claim

EvoStruct bridges a frozen protein language model with 3D structural context from an E(3)-equivariant GNN via a cross-attention adapter. Progressive PLM unfreezing and R-Drop consistency regularization are added specifically to counter vocabulary collapse in CDR design. On the CHIMERA-Bench dataset EvoStruct records the highest amino acid recovery and lowest perplexity among compared methods, improving sequence recovery by 16 percent and reducing perplexity by 43 percent relative to the strongest GNN baseline while also recovering 2.3 times greater amino acid diversity and the highest binding-pair correlation with ground truth.

What carries the argument

The cross-attention adapter that injects 3D structural features from the equivariant GNN into the frozen protein language model while enabling controlled progressive unfreezing of the language-model layers.

If this is right

  • Antibody design pipelines gain access to candidate CDRs with substantially higher sequence diversity while maintaining or improving recovery rates.
  • Purely structural models systematically under-represent residues that appear frequently in evolutionary alignments but rarely in the structural training set.
  • Consistency regularization during progressive unfreezing prevents the adapted language model from overfitting to the narrow distribution of the design benchmark.
  • Designed sequences show stronger correlation with experimentally observed binding pairs, suggesting improved functional relevance.

Where Pith is reading between the lines

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

  • The same adapter pattern may transfer to design tasks for other protein families where evolutionary sequence data is abundant but high-resolution structures remain limited.
  • Laboratory validation of the recovered sequences would test whether the measured increase in diversity produces a corresponding rise in successful binding affinities.
  • Extending the method from isolated CDRs to full antibody variable domains would require additional mechanisms to preserve inter-loop and inter-chain structural consistency.

Load-bearing premise

That GNN encoders discard evolutionary substitution patterns and that a cross-attention adapter plus progressive unfreezing and R-Drop regularization can restore those patterns without introducing new biases or overfitting to the CHIMERA-Bench distribution.

What would settle it

Training the same GNN architecture on the identical structural dataset but explicitly augmenting its input with multiple-sequence alignments from evolutionary databases and still observing vocabulary collapse would falsify the claim that evolutionary priors are the missing component.

Figures

Figures reproduced from arXiv: 2605.21485 by Mansoor Ahmed, Murray Patterson, Sujin Lee, Umar Khayaz.

Figure 1
Figure 1. Figure 1: EVOSTRUCT architecture. Two parallel paths process the antibody-antigen complex. A Relational-EGNN encoder produces structural embeddings over the full complex graph, while frozen ESM-2 produces evolutionary embeddings for CDR positions. The Mini Structural Adapter cross-attends ESM-2 CDR queries to GNN-encoded structural keys/values (CDR + nearest antigen residues), whereas the Sequence Head produces the … view at source ↗
Figure 2
Figure 2. Figure 2: Amino acid recovery (AAR) comparison. EVOSTRUCT (red) achieves 0.43 AAR, a 16% relative improvement over the best GNN baselines (0.37). Blue: GNN methods. Orange: diffu￾sion/flow. Green: ODE. Purple: autoregressive. with hidden dimension dgnn = 256, amino acid embedding dimension 32, and 8 edge types. We crop the Kag = 128 nearest antigen residues as structural context. The adapter operates in dimension da… view at source ↗
Figure 3
Figure 3. Figure 3: Vocabulary collapse. Per-position amino acid frequency heatmaps. EVOSTRUCT preserves near-native diversity (Veff=12.4). ward methods that excel in both sequence and structural quality. 5.4. Failure Mode Resolution We evaluate EVOSTRUCT on the three failure modes identi￾fied in Section 3.3. Vocabulary diversity. EVOSTRUCT achieves Veff = 12.4, recovering 80% of the ground-truth amino acid diversity ( [PITH… view at source ↗
Figure 4
Figure 4. Figure 4: Binding-pair preferences. Paratope-epitope AA pair heatmaps. EVOSTRUCT achieves the highest pair correlation (r=0.73) with ground truth, capturing binding-specific preferences [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Positional AAR profile. Per-position recovery along CDR-H3. EVOSTRUCT (red dashed) leads at both anchor positions and the hypervariable apex. 5.5. Discussion The results confirm that the PLM-structure adapter paradigm (Zheng et al., 2023), originally developed for gen￾eral protein inverse folding, transfers effectively to CDR de￾sign. The vocabulary calibration of ESM-2 transfers directly through the adapt… view at source ↗
Figure 6
Figure 6. Figure 6: Vocabulary collapse. Per-position amino acid frequency heatmaps across all the benchmarked baseline methods. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Binding-pair preferences. Paratope-epitope AA pair heatmaps across all the benchmarked baseline methods. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
read the original abstract

Equivariant graph neural network (GNN) methods for antibody complementarity-determining region (CDR) design achieve the highest sequence recovery but suffer from severe vocabulary collapse. The current best GNN methods over-predict very few amino acids, such as tyrosine and glycine, while ignoring functionally important residues. We trace this failure to GNN encoders learning amino acid distributions de novo from limited structural data, discarding substitution patterns encoded in evolutionary databases. To resolve this, we propose EvoStruct, which bridges a frozen protein language model (PLM) with 3D structural context from an E(3)-equivariant GNN via a cross-attention adapter. Unlike prior PLM-structure adapters for general protein design, EvoStruct targets the vocabulary collapse problem specific to CDR design through progressive PLM unfreezing and R-Drop consistency regularization. On the CHIMERA-Bench dataset, EvoStruct achieves the highest amino acid recovery and lowest perplexity among several antibody design methods, improving sequence recovery by 16% and reducing perplexity by 43% relative to the best GNN baselines, while recovering 2.3x greater amino acid diversity and the highest binding-pair correlation with ground truth.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces EvoStruct, which adapts a protein language model (PLM) to antibody CDR design by connecting it to an E(3)-equivariant GNN via a cross-attention adapter. Progressive unfreezing of the PLM and R-Drop regularization are used to mitigate vocabulary collapse (over-prediction of residues such as tyrosine and glycine) that the authors attribute to GNNs learning distributions de novo from limited structural data. On CHIMERA-Bench the method is reported to deliver the highest amino-acid recovery and lowest perplexity among compared antibody design approaches, with a 16% recovery gain and 43% perplexity reduction relative to the strongest GNN baseline, plus 2.3× greater diversity and the highest binding-pair correlation with ground truth.

Significance. If the gains are shown to arise specifically from the evolutionary substitution statistics supplied by the PLM rather than from the unfreezing schedule or R-Drop alone, the work would offer a concrete route to combining evolutionary and structural priors for a practically important design task. The targeted diagnosis of vocabulary collapse and the adapter-plus-regularization recipe are technically interesting; however, the current evidence does not yet isolate the contribution of the PLM, limiting the strength of the central bridging claim.

major comments (2)
  1. [Results section] Results (CHIMERA-Bench tables/figures): the abstract and reported numeric gains (16% recovery, 43% perplexity) supply no information on train/validation/test splits, number of independent runs, statistical significance tests, or error bars. Without these details the magnitude and reliability of the headline improvements cannot be assessed.
  2. [Ablation studies] Ablation studies (presumably §4 or equivalent): the manuscript does not report the control in which the PLM is replaced by a randomly initialized transformer while the cross-attention adapter, progressive unfreezing schedule, and R-Drop regularization are held fixed. This ablation is load-bearing for the claim that performance improvements derive from evolutionary substitution patterns encoded in the PLM rather than from the training techniques themselves.
minor comments (2)
  1. [Abstract] Abstract: the phrase 'several antibody design methods' is used without naming the baselines; an explicit list would improve readability.
  2. [Introduction] Notation: ensure that 'vocabulary collapse' is defined once and then used consistently; the current description mixes 'over-predict very few amino acids' with 'ignoring functionally important residues' without a quantitative definition.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting the need for clearer statistical reporting and a key ablation to strengthen the central claim. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Results section] Results (CHIMERA-Bench tables/figures): the abstract and reported numeric gains (16% recovery, 43% perplexity) supply no information on train/validation/test splits, number of independent runs, statistical significance tests, or error bars. Without these details the magnitude and reliability of the headline improvements cannot be assessed.

    Authors: We agree that additional details are required for rigorous assessment of the reported gains. In the revised manuscript we will explicitly describe the CHIMERA-Bench train/validation/test splits, report results aggregated over five independent random seeds with mean and standard deviation, add error bars to all relevant figures, and include paired statistical significance tests (e.g., Wilcoxon signed-rank) comparing EvoStruct against the strongest baselines. revision: yes

  2. Referee: [Ablation studies] Ablation studies (presumably §4 or equivalent): the manuscript does not report the control in which the PLM is replaced by a randomly initialized transformer while the cross-attention adapter, progressive unfreezing schedule, and R-Drop regularization are held fixed. This ablation is load-bearing for the claim that performance improvements derive from evolutionary substitution patterns encoded in the PLM rather than from the training techniques themselves.

    Authors: We acknowledge that this control experiment would provide the most direct evidence isolating the contribution of the pre-trained evolutionary priors. We will run the requested ablation (randomly initialized transformer with identical adapter, unfreezing schedule, and R-Drop) and report the results in the revised manuscript to quantify how much of the observed recovery and diversity gains are attributable to the PLM versus the training recipe alone. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on external benchmarks

full rationale

The paper hypothesizes that GNNs discard evolutionary substitution patterns due to limited structural data and proposes EvoStruct to address vocabulary collapse via a cross-attention adapter between a frozen PLM and E(3)-equivariant GNN, combined with progressive unfreezing and R-Drop regularization. All central claims are supported by direct empirical comparisons on the CHIMERA-Bench dataset, reporting improvements in amino acid recovery, perplexity, diversity, and binding-pair correlation relative to prior GNN baselines. No equations, derivations, or steps reduce a claimed result to a fitted parameter or self-citation by construction; the method components are presented as design choices, and performance metrics are externally falsifiable against the stated baselines without internal redefinition.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper rests on the domain assumption that vocabulary collapse originates from GNNs learning distributions de novo from limited structural data; no free parameters or invented physical entities are introduced in the abstract.

axioms (1)
  • domain assumption GNN encoders for CDR design learn amino acid distributions de novo from limited structural data and therefore discard substitution patterns present in evolutionary sequence databases.
    This premise is stated as the root cause of vocabulary collapse and motivates the entire adapter design.

pith-pipeline@v0.9.0 · 5751 in / 1441 out tokens · 30686 ms · 2026-05-21T05:02:27.287760+00:00 · methodology

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

300 extracted references · 300 canonical work pages · 5 internal anchors

  1. [1]

    2026 , url=

    Mansoor Ahmed and Nadeem Taj and Imdad Ullah Khan and Hemanth Venkateswara and Murray Patterson , booktitle=. 2026 , url=

    work page 2026

  2. [2]

    biorxiv , title =

    Chen, Xingyao and Dougherty, Thomas and Hong, Chan and Schibler, Rachel and Zhao, Yi and Sadeghi, Reza and Matasci, Naim and Wu, Yi-Chieh and Kerman, Ian , year =. biorxiv , title =

    work page

  3. [3]

    Prediction of Antibody-Antigen Binding via Machine Learning: Development of Data Sets and Evaluation of Methods

    Ye, Chao and Hu, Wenxing and Gaeta, Bruno. Prediction of Antibody-Antigen Binding via Machine Learning: Development of Data Sets and Evaluation of Methods. JMIR Bioinform Biotech. 2022. doi:10.2196/29404

    work page doi:10.2196/29404 2022

  4. [4]

    and Friedensohn, Simon and Weber, C

    Mason, Derek M. and Friedensohn, Simon and Weber, C. Optimization of therapeutic antibodies by predicting antigen specificity from antibody sequence via deep learning , journal=. 2021 , month=. doi:10.1038/s41551-021-00699-9 , url=

    work page doi:10.1038/s41551-021-00699-9 2021

  5. [5]

    MAbs , volume=

    Discovery-stage identification of drug-like antibodies using emerging experimental and computational methods , author=. MAbs , volume=. 2021 , organization=

    work page 2021

  6. [6]

    Antibody stability: A key to performance - Analysis, influences and improvement , journal =

    Hui Ma and Ciarán Ó’Fágáin and Richard O’Kennedy , keywords =. Antibody stability: A key to performance - Analysis, influences and improvement , journal =. 2020 , issn =. doi:https://doi.org/10.1016/j.biochi.2020.08.019 , url =

    work page doi:10.1016/j.biochi.2020.08.019 2020

  7. [7]

    doi:https://doi.org/10.1111/cbdd.13388 , url =

    Muhammed, Muhammed Tilahun and Aki-Yalcin, Esin , title =. doi:https://doi.org/10.1111/cbdd.13388 , url =. https://onlinelibrary.wiley.com/doi/pdf/10.1111/cbdd.13388 , year =

    work page doi:10.1111/cbdd.13388

  8. [8]

    Journal of Biomedical Science , year=

    Lu, Ruei-Min and Hwang, Yu-Chyi and Liu, I-Ju and Lee, Chi-Chiu and Tsai, Han-Zen and Li, Hsin-Jung and Wu, Han-Chung , title=. Journal of Biomedical Science , year=. doi:10.1186/s12929-019-0592-z , url=

    work page doi:10.1186/s12929-019-0592-z

  9. [9]

    Grange, R. D. and Thompson, J. P. and Lambert, D. G. , title = ". BJA: British Journal of Anaesthesia , volume =. 2014 , month =. doi:10.1093/bja/aet293 , url =

    work page doi:10.1093/bja/aet293 2014

  10. [10]

    Frontiers in Immunology , VOLUME=

    Huang, Yan and Zhang, Ziding and Zhou, Yuan , TITLE=. Frontiers in Immunology , VOLUME=. 2022 , URL=. doi:10.3389/fimmu.2022.1053617 , ISSN=

    work page doi:10.3389/fimmu.2022.1053617 2022

  11. [11]

    PLOS ONE , publisher =

    Assisted Design of Antibody and Protein Therapeutics (ADAPT) , year =. PLOS ONE , publisher =. doi:10.1371/journal.pone.0181490 , author =

    work page doi:10.1371/journal.pone.0181490

  12. [12]

    and Ascher, David B

    Pires, Douglas E.V. and Ascher, David B. , title = ". Nucleic Acids Research , volume =. 2016 , month =. doi:10.1093/nar/gkw458 , url =

    work page doi:10.1093/nar/gkw458 2016

  13. [13]

    Khamis and Walid Gomaa and Walaa F

    Mohamed A. Khamis and Walid Gomaa and Walaa F. Ahmed , keywords =. Machine learning in computational docking , journal =. 2015 , issn =. doi:https://doi.org/10.1016/j.artmed.2015.02.002 , url =

    work page doi:10.1016/j.artmed.2015.02.002 2015

  14. [14]

    and Andersen, Jan Terje and Greiff, Victor , title=

    Akbar, Rahmad and Bashour, Habib and Rawat, Puneet and Robert, Philippe A. and Andersen, Jan Terje and Greiff, Victor , title=. mAbs , year=. doi:10.1080/19420862.2021.2008790 , url=

    work page doi:10.1080/19420862.2021.2008790 2021

  15. [15]

    and Lehmann, Paul V

    Ras-Carmona, Alvaro and Lehmann, Alexander A. and Lehmann, Paul V. and Reche, Pedro A. , title=. Scientific Reports , year=. doi:10.1038/s41598-022-18021-1 , url=

    work page doi:10.1038/s41598-022-18021-1

  16. [16]

    BMC Genomics , year=

    Ren, Jing and Song, Jiangning and Ellis, John and Li, Jinyan , title=. BMC Genomics , year=. doi:10.1186/s12864-017-3493-0 , url=

    work page doi:10.1186/s12864-017-3493-0

  17. [17]

    BepFAMN : A Method for Linear B-Cell Epitope Predictions Based on Fuzzy-ARTMAP Artificial Neural Network

    La Marca, Anthony F and Lopes, Robson da S and Lotufo, Anna Diva P and Bartholomeu, Daniella C and Minussi, Carlos R. BepFAMN : A Method for Linear B-Cell Epitope Predictions Based on Fuzzy-ARTMAP Artificial Neural Network. Sensors (Basel)

    work page

  18. [18]

    Bioinformatics , volume =

    Liberis, Edgar and Veličković, Petar and Sormanni, Pietro and Vendruscolo, Michele and Liò, Pietro , title = ". Bioinformatics , volume =. 2018 , month =. doi:10.1093/bioinformatics/bty305 , url =

    work page doi:10.1093/bioinformatics/bty305 2018

  19. [19]

    Bioinformatics , volume=

    Paragraph—antibody paratope prediction using graph neural networks with minimal feature vectors , author=. Bioinformatics , volume=. 2023 , publisher=

    work page 2023

  20. [20]

    2025 , doi =

    Ahmed, Mansoor and Ali, Sarwan and Jan, Avais and Khan, Imdad Ullah and Patterson, Murray , title =. 2025 , doi =

    work page 2025

  21. [21]

    Antibodies , volume=

    In silico methods in antibody design , author=. Antibodies , volume=. 2018 , publisher=

    work page 2018

  22. [22]

    Mabs , volume=

    Artificial intelligence-driven computational methods for antibody design and optimization , author=. Mabs , volume=. 2025 , organization=

    work page 2025

  23. [23]

    arXiv preprint arXiv:2506.04235 , year=

    AbBiBench: A Benchmark for Antibody Binding Affinity Maturation and Design , author=. arXiv preprint arXiv:2506.04235 , year=

    work page arXiv

  24. [24]

    PloS one , volume=

    OptMAVEn--a new framework for the de novo design of antibody variable region models targeting specific antigen epitopes , author=. PloS one , volume=. 2014 , publisher=

    work page 2014

  25. [25]

    Attentive Cross-Modal Paratope Prediction , journal =

    Deac, Andreea and Veli. Attentive Cross-Modal Paratope Prediction , journal =. 2019 , doi =

    work page 2019

  26. [26]

    Nature Biotechnology , year=

    A trimodal protein language model enables advanced protein searches , author=. Nature Biotechnology , year=. doi:10.1038/s41587-025-02836-0 , url=

    work page doi:10.1038/s41587-025-02836-0

  27. [27]

    Nucleic acids research , year=

    Chailyan, Anna and Tramontano, Anna and Marcatili, Paolo , title=. Nucleic acids research , year=. doi:10.1093/nar/gkr806 , url=

    work page doi:10.1093/nar/gkr806

  28. [28]

    and Johnson, David S

    Lim, Yoong Wearn and Adler, Adam S. and Johnson, David S. , title=. mAbs , year=. doi:10.1080/19420862.2022.2069075 , url=

    work page doi:10.1080/19420862.2022.2069075 2022

  29. [29]

    Briefings in bioinformatics , volume=

    AntiFormer: graph enhanced large language model for binding affinity prediction , author=. Briefings in bioinformatics , volume=. 2024 , publisher=

    work page 2024

  30. [30]

    Bioinformatics , volume=

    CSM-AB: Graph-based antibody--antigen binding affinity prediction and docking scoring function , author=. Bioinformatics , volume=. 2022 , publisher=

    work page 2022

  31. [31]

    Structure , volume=

    ANTIPASTI: interpretable prediction of antibody binding affinity exploiting Normal Modes and Deep Learning , author=. Structure , volume=. 2024 , publisher=

    work page 2024

  32. [32]

    Nature Machine Intelligence , volume=

    A topology-based network tree for the prediction of protein--protein binding affinity changes following mutation , author=. Nature Machine Intelligence , volume=. 2020 , publisher=

    work page 2020

  33. [33]

    Journal of Computational Biology , volume=

    Reads2vec: Efficient embedding of raw high-throughput sequencing reads data , author=. Journal of Computational Biology , volume=. 2023 , publisher=

    work page 2023

  34. [34]

    Algorithms , volume=

    Robust representation and efficient feature selection allows for effective clustering of sars-cov-2 variants , author=. Algorithms , volume=. 2021 , publisher=

    work page 2021

  35. [35]

    Bioinformatics Research and Applications: 17th International Symposium, ISBRA 2021, Shenzhen, China, November 26--28, 2021, Proceedings 17 , pages=

    A k-mer based approach for sars-cov-2 variant identification , author=. Bioinformatics Research and Applications: 17th International Symposium, ISBRA 2021, Shenzhen, China, November 26--28, 2021, Proceedings 17 , pages=. 2021 , organization=

    work page 2021

  36. [36]

    Biology , volume=

    Exploring the Potential of GANs in Biological Sequence Analysis , author=. Biology , volume=. 2023 , publisher=

    work page 2023

  37. [37]

    Bioinformatics , volume=

    DLAB: deep learning methods for structure-based virtual screening of antibodies , author=. Bioinformatics , volume=. 2022 , publisher=

    work page 2022

  38. [38]

    Bioinformatics , volume =

    Myung, Yoochan and Pires, Douglas E V and Ascher, David B , title = ". Bioinformatics , volume =. 2021 , month =. doi:10.1093/bioinformatics/btab762 , url =

    work page doi:10.1093/bioinformatics/btab762 2021

  39. [39]

    Machine learning prediction of Antibody-Antigen binding: dataset, method and testing , journal=

    Ye, Chao and Hu, Wenxing and Gaëta, Bruno , year =. Machine learning prediction of Antibody-Antigen binding: dataset, method and testing , journal=

    work page

  40. [40]

    Bioinformatics , volume =

    Jain, Tushar and Boland, Todd and Lilov, Asparouh and Burnina, Irina and Brown, Michael and Xu, Yingda and Vásquez, Maximiliano , title = ". Bioinformatics , volume =. 2017 , month =. doi:10.1093/bioinformatics/btx519 , url =

    work page doi:10.1093/bioinformatics/btx519 2017

  41. [41]

    Frontiers in Microbiology , VOLUME=

    Kang, Tae Hyun and Seong, Baik Lin , TITLE=. Frontiers in Microbiology , VOLUME=. 2020 , URL=. doi:10.3389/fmicb.2020.01927 , ISSN=

    work page doi:10.3389/fmicb.2020.01927 2020

  42. [42]

    Antibody apparent solubility prediction from sequence by transfer learning , journal =

    Jiangyan Feng and Min Jiang and James Shih and Qing Chai , keywords =. Antibody apparent solubility prediction from sequence by transfer learning , journal =. 2022 , issn =. doi:https://doi.org/10.1016/j.isci.2022.105173 , url =

    work page doi:10.1016/j.isci.2022.105173 2022

  43. [43]

    Alejandro and Charonis, Spyros and Curtis, Robin and Warwicker, Jim , title =

    Hebditch, Max and Carballo-Amador, M. Alejandro and Charonis, Spyros and Curtis, Robin and Warwicker, Jim , title =. 2017 , journal =. doi:10.1093/bioinformatics/btx345 , url =

    work page doi:10.1093/bioinformatics/btx345 2017

  44. [44]

    Computational and artificial intelligence-based methods for antibody development

    Kim, Jisun and McFee, Matthew and Fang, Qiao and Abdin, Osama and Kim, Philip M. Computational and artificial intelligence-based methods for antibody development. Trends Pharmacol Sci

    work page

  45. [45]

    Frontiers in immunology , volume=

    DiscoTope-3.0: improved B-cell epitope prediction using inverse folding latent representations , author=. Frontiers in immunology , volume=. 2024 , publisher=

    work page 2024

  46. [46]

    Antibody Therapeutics , volume =

    Zhang, Weijie and Wang, Hao and Feng, Nan and Li, Yifeng and Gu, Jijie and Wang, Zhuozhi , title = ". Antibody Therapeutics , volume =. 2022 , month =. doi:10.1093/abt/tbac029 , url =

    work page doi:10.1093/abt/tbac029 2022

  47. [47]

    1996 , issn =

    Computational methods for biomolecular docking , journal =. 1996 , issn =. doi:https://doi.org/10.1016/S0959-440X(96)80061-3 , url =

    work page doi:10.1016/s0959-440x(96)80061-3 1996

  48. [48]

    Protein structure prediction

    Deng, Haiyou and Jia, Ya and Zhang, Yang. Protein structure prediction. Int. J. Mod. Phys. B

    work page

  49. [49]

    MAbs , volume=

    Antibodies to watch in 2019 , author=. MAbs , volume=. 2019 , organization=

    work page 2019

  50. [50]

    MAbs , volume=

    Phage display and hybridoma generation of antibodies to human CXCR2 yields antibodies with distinct mechanisms and epitopes , author=. MAbs , volume=. 2014 , organization=

    work page 2014

  51. [51]

    Immunology today , volume=

    Antibodies in diagnostics--from immunoassays to protein chips , author=. Immunology today , volume=. 2000 , publisher=

    work page 2000

  52. [52]

    Nature , volume=

    Highly accurate protein structure prediction with AlphaFold , author=. Nature , volume=. 2021 , publisher=

    work page 2021

  53. [53]

    Education, the Responsible and Ethical Use of ChatGPT Towards Lifelong Learning (February 11, 2023) , year=

    Open AI in education, the responsible and ethical use of ChatGPT towards lifelong learning , author=. Education, the Responsible and Ethical Use of ChatGPT Towards Lifelong Learning (February 11, 2023) , year=

    work page 2023

  54. [54]

    and Vendruscolo, Michele

    Sormanni, Pietro and Aprile, Francesco A. and Vendruscolo, Michele. Third generation antibody discovery methods: in silico rational design. Chem. Soc. Rev. 2018. doi:10.1039/C8CS00523K

    work page doi:10.1039/c8cs00523k 2018

  55. [55]

    Nucleic acids research , volume=

    PyIgClassify: a database of antibody CDR structural classifications , author=. Nucleic acids research , volume=. 2015 , publisher=

    work page 2015

  56. [56]

    Current Opinion in Structural Biology , volume=

    Advances in computational structure-based antibody design , author=. Current Opinion in Structural Biology , volume=. 2022 , publisher=

    work page 2022

  57. [57]

    Bioinformatics , volume =

    Dai, Bowen and Bailey-Kellogg, Chris , title = ". Bioinformatics , volume =. 2021 , month =

    work page 2021

  58. [58]

    Scientific reports , volume=

    Automation of absolute protein-ligand binding free energy calculations for docking refinement and compound evaluation , author=. Scientific reports , volume=. 2021 , publisher=

    work page 2021

  59. [59]

    bioRxiv , pages=

    Advancing Protein-DNA Binding Site Prediction: Integrating Sequence Models and Machine Learning Classifiers , author=. bioRxiv , pages=. 2023 , publisher=

    work page 2023

  60. [60]

    Current opinion in virology , volume=

    Antibody specific epitope prediction—emergence of a new paradigm , author=. Current opinion in virology , volume=. 2015 , publisher=

    work page 2015

  61. [61]

    2024 , publisher=

    Wang, Chuan and Wang, Jiangyuan and Song, Wenjun and Luo, Guanzheng and Jiang, Taijiao , journal=. 2024 , publisher=

    work page 2024

  62. [62]

    epitope1D: accurate taxonomy-aware

    Silva, Bruna and Ascher, David and Pires, Douglas , journal=. epitope1D: accurate taxonomy-aware. 2023 , publisher=

    work page 2023

  63. [63]

    Bioinformatics , volume=

    Learning context-aware structural representations to predict antigen and antibody binding interfaces , author=. Bioinformatics , volume=. 2020 , publisher=

    work page 2020

  64. [64]

    Liu, Chunan and Denzler, Lilian and Chen, Yihong and Martin, Andrew and Paige, Brooks , journal=

    work page

  65. [65]

    A structure-based

    Lu, Shuai and Li, Yuguang and Ma, Qiang and Nan, Xiaofei and Zhang, Shoutao , journal=. A structure-based. 2022 , publisher=

    work page 2022

  66. [66]

    Scientific Reports , volume=

    Prediction of protein--protein interaction using graph neural networks , author=. Scientific Reports , volume=. 2022 , publisher=

    work page 2022

  67. [67]

    Frontiers in immunology , volume=

    SEMA: Antigen B-cell conformational epitope prediction using deep transfer learning , author=. Frontiers in immunology , volume=. 2022 , publisher=

    work page 2022

  68. [68]

    Nucleic Acids Research , volume=

    SEPPA-mAb: spatial epitope prediction of protein antigens for mAbs , author=. Nucleic Acids Research , volume=. 2023 , publisher=

    work page 2023

  69. [69]

    Briefings in bioinformatics , volume=

    Critical review of conformational B-cell epitope prediction methods , author=. Briefings in bioinformatics , volume=. 2023 , publisher=

    work page 2023

  70. [70]

    Nature protocols , volume=

    The ClusPro AbEMap web server for the prediction of antibody epitopes , author=. Nature protocols , volume=. 2023 , publisher=

    work page 2023

  71. [71]

    Journal of Biological Chemistry , volume=

    Cloning and characterization of deoxymugineic acid synthase genes from graminaceous plants , author=. Journal of Biological Chemistry , volume=. 2006 , publisher=

    work page 2006

  72. [72]

    coli , author=

    Small ubiquitin-like modifier protein 3 enhances the solubilization of human bone morphogenetic protein 2 in E. coli , author=. Applied biochemistry and biotechnology , volume=. 2018 , publisher=

    work page 2018

  73. [73]

    Journal of Experimental Botany , volume=

    Paralogs and mutants show that one DMA synthase functions in iron homeostasis in rice , author=. Journal of Experimental Botany , volume=. 2017 , publisher=

    work page 2017

  74. [74]

    , author=

    Effects of MDL 19205 (piroximone), a new cardiotonic agent, on electrophysiological, mechanical, and intracellular ionic characteristics of sheep cardiac tissues. , author=. Journal of cardiovascular pharmacology , volume=

    work page

  75. [75]

    Nature communications , volume=

    Histone H4 lysine 20 mono-methylation directly facilitates chromatin openness and promotes transcription of housekeeping genes , author=. Nature communications , volume=. 2021 , publisher=

    work page 2021

  76. [76]

    Nature communications , volume=

    Histone H4K20 methylation mediated chromatin compaction threshold ensures genome integrity by limiting DNA replication licensing , author=. Nature communications , volume=. 2018 , publisher=

    work page 2018

  77. [77]

    Plant molecular biology , volume=

    Iron deficiency regulated OsOPT7 is essential for iron homeostasis in rice , author=. Plant molecular biology , volume=. 2015 , publisher=

    work page 2015

  78. [78]

    Nucleic Acids Research , volume=

    Topokaryotyping demonstrates single cell variability and stress dependent variations in nuclear envelope associated domains , author=. Nucleic Acids Research , volume=. 2018 , publisher=

    work page 2018

  79. [79]

    in vivo biotinylation

    PUB-NChIP—“in vivo biotinylation” approach to study chromatin in proximity to a protein of interest , author=. Genome research , volume=. 2013 , publisher=

    work page 2013

  80. [80]

    Journal of proteome research , volume=

    PUB-MS: a mass spectrometry-based method to monitor protein--protein proximity in vivo , author=. Journal of proteome research , volume=. 2011 , publisher=

    work page 2011

Showing first 80 references.