Fast and Accurate Antibody Sequence Design via Structure Retrieval

Biaobin Jiang; Kangfei Zhao; Kun Xie; Ningqiao Huang; Peilin Zhao; Sibo Wang; Wei Liu; Xingyi Zhang

REVIEW 4 major objections 5 minor 1 cited by

Fast and Accurate Antibody Sequence Design via Structure Retrieval

T0 review · 4 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read IgSeek establishes that antibody CDR sequences can be designed by retrieving structurally similar natural loops from a vector database, outperforming inverse-folding models on sequence recovery while reducing inference cost by at least 20x.

desk verdict A clever retrieval-based CDR design idea whose headline numbers are undercut by unfair baselines and unquantified database redundancy. read the letter →

arxiv 2502.19395 v1 pith:AMAOEW2R submitted 2025-02-11 q-bio.BM cs.LG

Xingyi Zhang , Kun Xie , Ningqiao Huang , Wei Liu , Peilin Zhao , Sibo Wang , Kangfei Zhao , Biaobin Jiang This is my paper · ORCID

classification q-bio.BMcs.LG

keywords antibodydesignCDRsequencestructureretrievalinversefoldingequivariantgraphneuralnetworkrecoveryT-cellreceptorhallucinationreduction

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

Antibodies bind their targets through hypervariable loops called CDRs, and the standard deep-learning route designs them by inverse folding: predicting a sequence from a generated backbone. This paper argues that a simpler route works better for CDRs—retrieve the most structurally similar loops from a database of natural antibodies and assemble a sequence from their conserved motifs. The proposed framework, IgSeek, embeds 24,479 solved antibody CDR structures into a vector space, then for any query loop pulls the ten nearest equal-length neighbors and samples each residue from the aligned sequences of those neighbors. On antibodies released in 2024, IgSeek recovers at least 2.9% more native amino acids than state-of-the-art inverse-folding models on light-chain CDRs, transfers to T-cell receptor CDRs without retraining, and is at least 20x faster at inference. If right, this reframes antibody sequence design as a database coverage problem rather than a generative modeling problem, and it directly reduces the hallucination risk that plagues generative inverse folding.

What carries the argument

The load-bearing object is MEGNN (Multi-channel Equivariant Graph Neural Network), an E(3)-equivariant encoder that represents each residue by its four backbone atoms as separate channels and reads out a fixed-length embedding per CDR. It is trained self-supervised to predict the RMSD of CDR pairs whose TM-score is over 0.6, so embeddings live in a space where distance means structural similarity. At inference, the query embedding is the key for a K-NN search over 24,479 solved CDR loops from SAbDab-before-2024; the top-10 nearest equal-length CDRs contribute their aligned sequences as an empirical amino-acid frequency at each loop position, from which the final sequence is sampled. An optional variant, IgSeek+Kabsch, re-ranks the retrieved candidates by exact RMSD to simulate perfect retrieval and shows the headroom in the recovery numbers.

What would settle it

Run IgSeek on a test set of CDR backbones that are deliberately shifted so that no database loop is within 2 Å RMSD; if the average amino-acid recovery does not drop sharply toward random, the method's gains are not actually coming from retrieved structural templates.

Watch

Extended reading notes

Core claim

IgSeek's central discovery is that CDR sequence inference can be posed as isomorphic structure retrieval: an E(3)-equivariant multi-channel graph neural network (MEGNN) encodes each CDR backbone into a fixed-length vector, trained self-supervised so that embedding distance predicts pairwise backbone RMSD; then the query embedding retrieves top-10 equal-length nearest neighbors from a vector database of 24,479 solved CDR loops from before 2024. The predicted sequence is generated position by position by counting amino acids at aligned positions of the retrieved sequences and sampling from that empirical distribution. The recovery numbers on the 4,449 SAbDab-2024 CDRs are the claim: at least 2.9% higher average amino-acid recovery than ProteinMPNN, ESM-IF1, AbMPNN, and AntiFold on light-chain CDRs, comparable recovery on heavy-chain CDRs, consistent wins over all baselines on all six CDR types when the retrieved candidates are re-ranked by exact RMSD (IgSeek+Kabsch), at least 30% higher recovery on T-cell receptor CDRs without retraining, and at least 20x faster inference than the generative baselines.

Load-bearing premise

The method assumes the database holds, for every query CDR of a given length, a few loops whose backbone shapes are close enough (roughly within 1 Å RMSD) that their aligned sequence motifs can be recombined to recover the native sequence; if a query has no such close structural neighbor, retrieval has no information to contribute.

Editorial extensions

If this is right

Larger retrieval databases should translate directly into higher sequence recovery, because every added natural loop expands the set of structural templates IgSeek can draw from.
IgSeek can replace the inverse-folding step in existing antibody structure-generation pipelines; in the paper's 8R1C case study, RFdiffusion-plus-IgSeek recovered 0.67 of native residues at 0.38 Å backbone RMSD, versus 0.33 for dyMEAN alone.
Because IgSeek needs only the CDR backbone and no framework sequence, it can design CDR sequences on any fixed antibody framework, which is the standard developability-friendly setup.
The 20x inference speed-up makes high-throughput design over large repertoires (the paper tests 5,000 predicted CDR-H3 loops) practical without retraining.

Reading between the lines

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

A direct test of the hallucination-reduction claim that the paper does not report: fold IgSeek-generated sequences with an independent structure predictor and compare CDR backbone RMSD to the query structure; retrieval should visibly suffer less distribution shift than inverse-folding baselines.
The dependence on database coverage suggests an experiment the authors did not run—construct a retrieval database augmented with high-confidence predicted antibody structures and measure whether long, rare CDR-H3 loops, where solved examples are scarce, gain the most.
Position-wise sampling discards correlations between residue choices within a loop; modeling the joint distribution over aligned retrieved motifs could improve both recovery and naturalness while keeping the retrieval mechanism.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

You should know two things about IgSeek (arXiv:2502.19395). First, it's a genuinely new angle on antibody CDR sequence design: instead of inverse folding with a learned generative model, it retrieves similar CDR structures from SAbDab and assembles the sequence as a consensus of the retrieved templates. That's clean, interpretable, and fast, and it sidesteps hallucination by construction. The MEGNN embedding that maps CDR backbones to vectors, trained to predict pairwise RMSD, is sensible, and the retrieval comparison against FoldSeek is a real, positive result.

Second, the headline claim—that IgSeek beats state-of-the-art inverse folding on AAR—is not yet supported. The main benchmark gives the baselines only the isolated CDR backbone, while AntiFold and AbMPNN expect the full variable domain. The authors acknowledge this in a Remark, but they don't fix it in the main results. The appendix adds flanking residues, but even then the baselines are not given their native input format.

The bigger issue is the one the stress-test note flags: IgSeek's accuracy depends almost entirely on whether the database contains a near-structural homolog of the query. The paper removes exact duplicates and uses a temporal split, which is good, but it only reports mean sequence similarity of 0.3–0.5 between train and test. That average hides the tail. If a substantial fraction of test CDRs have a near-duplicate in the retrieval database, the consensus method trivially recovers those sequences, while the inverse folding baselines get no such shortcut. Without stratifying AAR by distance to the nearest database neighbor or reporting the max-sequence-identity distribution for each query, the 2.9% lead on light chains could reflect database coverage rather than generalizable design ability. That's a load-bearing omission.

There are smaller soft spots: no error bars, no code or data, and no non-parametric retrieval baseline (e.g., TM-align or FoldSeek) used directly for sequence assembly in the AAR comparison. The FoldSeek comparison is only for retrieval quality, not for end-to-end AAR.

If the authors add the stratification and a fair baseline setting, this could be a solid contribution. As it stands, it's a promising idea with an evaluation that overclaims. I'd still send it to peer review—the idea is novel enough and the retrieval result interesting enough to warrant referee time to push on exactly these questions.

Referee Report

4 major / 5 minor

Summary. The paper proposes IgSeek, a retrieval-based method for antibody and T-cell receptor CDR sequence design. A multi-channel equivariant graph neural network (MEGNN) is pre-trained on SAbDab CDR pairs to predict pairwise RMSD (Eq. 14), then used to embed all CDRs in a pre-2024 SAbDab database. For a query CDR backbone, IgSeek retrieves K equal-length nearest neighbors in embedding space and samples each position from the empirical amino-acid histogram of the retrieved sequences. The authors evaluate amino-acid recovery (AAR) on SAbDab-2024 and STCRDab against ProteinMPNN, ESM-IF1, AbMPNN, and AntiFold, report order-of-magnitude speedups, and include a variant (IgSeek+Kabsch) that re-ranks retrieved candidates by true RMSD. The central claim is that retrieval-based inference outperforms inverse-folding baselines on CDR sequence recovery, especially for light-chain CDRs and T-cell receptors.

Significance. If the central comparison were robust, IgSeek would offer a useful and fast alternative to inverse folding for CDR design, and the temporal split is a commendable design choice. The paper also contains a formal equivariance proof and a clear algorithmic description. However, the current evidence is not sufficient to establish the headline advantage: the main benchmark gives baselines only CDR backbones, the STCRDab generalization test uses a random split that permits near-duplicate retrieval, and the database-coverage assumption is not quantified. These are correctable with additional experiments, so the work has potential but needs major revision.

major comments (4)

[§2.3 (Fig. 2a), Remark] The central comparison is not apples-to-apples. The authors state in §2.3 that only CDR backbone structures are given to all methods, and the Remark confirms that AntiFold and AbMPNN degrade because they rely on full antibody backbone and framework sequence information. IgSeek is specifically designed for CDR-only input, so part of its advantage may reflect task adaptation rather than design quality. The extended-context experiments in Appendix G (Fig. 8) add only up to three flanking residues, still not the full variable domain required by AntiFold. Moreover, the Abstract's claim that IgSeek 'outperforms state-of-the-art approaches' is stronger than what Fig. 2a shows for the primary variant: on heavy-chain CDRs IgSeek is comparable to baselines, and only IgSeek+Kabsch, which uses true RMSD for re-ranking, exceeds all baselines. Please re-run AntiFold and AbMPNN under their native input requirements, or clearly define the task as CDR-only and adapt the baselines accordingly, and make the abstract reflect the results of the primary variant.
[§2.3, Appendix D (Fig. 7)] The temporal split does not control for near-duplicate CDRs between the retrieval database and the test set. Appendix D says identical sequences are removed, but Fig. 7 reports only average sequence similarity of 0.3–0.5; an average does not rule out a long tail of near-duplicates, and for a retrieval method the tail is what matters. IgSeek's prediction is literally a consensus of the retrieved sequences, so a query with a close database homolog can be recovered trivially, whereas the inverse-folding baselines must infer the sequence from structure alone. The paper does not report the distribution of maximum sequence identity or minimum structural RMSD between each query and database entries, nor AAR stratified by these distances. Please add these stratifications and include a non-learned retrieval baseline (e.g., nearest neighbor by TM-align RMSD) to separate database-coverage effects from MEGNN embedding quality.
[§2.3 Generalization Performance, Appendix D (STCRDab)] The STCRDab generalization experiment uses a random 80/20 split of the same database to form templates and queries. Under a random split, many query CDRs will have close structural and sequence neighbors among the templates, so the reported ≥30% AAR lead over inverse-folding baselines in Fig. 2b is expected for a retrieval method and does not demonstrate generalization to genuinely novel receptors. Please use a temporal or cluster-based split, remove near-identical sequences between template and query sets, and report AAR after such filtering. Without this, the STCRDab result does not support the generalization claim made in the text.
[Fig. 2, Table 3, Table 5] No error bars, confidence intervals, or significance tests are reported for any AAR comparison. The headline difference on light-chain CDRs is 2.9 percentage points, and generation is stochastic (n_sample=2 in Table 3), so run-to-run variability and model-seed variability could change the ranking. The authors also report 'best results' for baselines but do not describe how many samples were used for each method. Please provide mean ± standard deviation over at least five runs and a paired significance test (e.g., Wilcoxon signed-rank) across CDR types to establish that the observed differences are not noise.

minor comments (5)

[§C.2, Algorithm 3] The procedure is described as 'Bernoulli sampling,' but sampling an amino acid from a 20-category probability distribution is categorical/multinomial sampling, not Bernoulli sampling. Please rename.
[§C.3, Theorem 1] The statement of Theorem 1 is malformed: the notation 'hi, TY(g)X_i^(L) = MEGNN(...)' mixes an equivariance equation with a tuple in an unclear way. Please rewrite the theorem statement with clear domain and codomain notation.
[Appendix D] The sentence 'we remove CDR sequences that are identical to those in the dataset to eliminate redundancy' is ambiguous: it is unclear whether removal is from the training set, the retrieval database, or the test set. Please clarify the exact filtering step and report how many CDRs were removed.
[Fig. 2c] The inference-time comparison on OAS-H3 should specify what exactly is timed for each method: IgSeek timing appears to be retrieval-plus-sampling only, while baselines include neural network inference; please define the measurement protocol and hardware.
[§3] There is a typo in the Conclusion: 'investigatino' should be 'investigation.'

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: sequence prediction is a K-NN consensus over retrieved CDR sequences, MEGNN regresses RMSD, and the test set is temporally separated.

full rationale

No circular step is present. IgSeek's predicted sequence is obtained by retrieving K equal-length CDR loops from a vector database and sampling from the per-position histogram of the retrieved sequences (Appendix C.2); the native sequence of the query is not used to construct the retrieval, the histogram, or the MEGNN parameters. MEGNN is trained to regress the RMSD between pairs of CDR structures (Eq. 14), not to maximize amino-acid recovery or to memorize query sequences, so the representation learning step is independent of the sequence-prediction objective. The test split (SAbDab-2024) is temporally disjoint from the retrieval database (SAbDab-before-2024), the paper states that it removes identical CDR sequences from the pre-2024 set and redundant loops from the test set, and the reported train/test sequence similarity is 0.3-0.5, so there is no equation-level or construction-level reduction of the target to an input. The main substantive concern is benchmark-level rather than circular: if a query CDR has a near-duplicate structural neighbor in the database, retrieval can recover its sequence by template consensus, and the paper does not stratify AAR by nearest-neighbor structural distance. That is a data-redundancy and generalizability risk, not a case in which the prediction is equivalent to an input by definition. The only self-citation (IgGM, via coauthor Peilin Zhao) appears in related work and is not load-bearing to any claimed result.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the empirical assumption that structure determines sequence at CDR positions, plus the coverage of the retrieval database. The method itself introduces no new biological entities; the MEGNN is a conventional equivariant network.

free parameters (4)

K (number of retrieved neighbors) = 10
Hand-chosen; Table 5 shows AAR declines with larger K, and the authors pick 10 as an accuracy-diversity trade-off.
TM-score threshold for training pairs = 0.6
Used to define structurally similar pairs for self-supervised training; chosen without sensitivity analysis.
Graph distance threshold theta = 10 Å
Neighborhood cutoff for graph construction; standard but hand-set.
Input noise ratio and scale for MEGNN = 0.15, sigma=1
Perturbation added to coordinates during training; chosen to improve robustness but not ablated.

assumptions (3)

domain assumption CDR backbone structure is a strong determinant of CDR sequence (canonical structure hypothesis)
The method's core premise, inherited from Chothia et al. (1989), that structurally similar CDRs share conserved sequence motifs. Invoked throughout and explicitly in Section 1.
domain assumption The SAbDab-before-2024 database has sufficient coverage of CDR structural space for retrieval to find useful templates
If the database lacks close structural homologs for a query, the K-NN search returns poor templates and sequence recovery collapses. The paper does not report minimum RMSD to database for test queries.
standard math E(3) equivariance of MEGNN is sufficient to encode structure in a retrieval-compatible embedding
The proof in Appendix C.3 establishes equivariance, but the quality of the embedding for K-NN search is only validated empirically via AUROC.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Fast and Accurate Antibody Sequence Design via Structure Retrieval." pith.science (2026). https://pith.science/paper/AMAOEW2R

@misc{pith2026250219395,
  author       = {Pith},
  title        = {Pith review of: Fast and Accurate Antibody Sequence Design via Structure Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AMAOEW2R}},
  note         = {Machine review of arXiv:2502.19395}
}

read the original abstract

Recent advancements in protein design have leveraged diffusion models to generate structural scaffolds, followed by a process known as protein inverse folding, which involves sequence inference on these scaffolds. However, these methodologies face significant challenges when applied to hyper-variable structures such as antibody Complementarity-Determining Regions (CDRs), where sequence inference frequently results in non-functional sequences due to hallucinations. Distinguished from prevailing protein inverse folding approaches, this paper introduces Igseek, a novel structure-retrieval framework that infers CDR sequences by retrieving similar structures from a natural antibody database. Specifically, Igseek employs a simple yet effective multi-channel equivariant graph neural network to generate high-quality geometric representations of CDR backbone structures. Subsequently, it aligns sequences of structurally similar CDRs and utilizes structurally conserved sequence motifs to enhance inference accuracy. Our experiments demonstrate that Igseek not only proves to be highly efficient in structural retrieval but also outperforms state-of-the-art approaches in sequence recovery for both antibodies and T-Cell Receptors, offering a new retrieval-based perspective for therapeutic protein design.

Figures

Figures reproduced from arXiv: 2502.19395 by the authors.

**Figure 1.** The Framework IgSeek: (a) Pre-train an MEGNN encoder by a self-supervised learning task. (b) Construct a CDR vector database. (c) Sequence generation by K-NN search. To deal with the challenge of hallucinations arising from previous models, we propose an antibody CDR sequence design framework from a novel perspective of similar structure retrieval, named as IgSeek (Ig for Immunoglobulin, a.k.a. antibody). Our fram… view at source ↗

**Figure 2.** The comparison of average AAR and inference speed. (a) AAR in SAbDab-2024 dataset. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** IgSeek vs. FoldSeek in CDR retrieval. In this set of experiments, we compare IgSeek with the state-of-theart structure searching model, FoldSeek Van Kempen et al. (2024), by examining the quality of the retrieved isomorphic structures. Introduction to competitors is deferred to Appendix F. Specifically, for a given query CDR q, the retrieved CDR r is considered a positive instance if their RMSD is less than 1 A.… view at source ↗

Figures from the paper (5 more)

**Figure 4.** Figure 4: Embeddings of CDRs in the SAbDab-before-2024 datasets projected onto 2D Space. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 5.** Figure 5: A Case study using 8W8R CDR-L1 as an example. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

**Figure 6.** Figure 6: Antibody Structure while the CDRs form the antigen-binding interface and are primarily responsible for the specificity and affinity of antigen recognition. These hypervariable CDR loops represent the most critical determinants of antibody-antigen interactions and are …

**Figure 7.** Figure 7: Sequence similarity between SAbDab train/test set. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

**Figure 8.** Figure 8: The Comparison of Average AAR on the SAbDab-2024 Dataset using CDRs with exten [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

Tokenizing Loops of Antibodies
q-bio.BM 2025-09 conditional novelty 6.0 of 10

Igloo is a multimodal antibody loop tokenizer that, when plugged into protein language models, modestly improves loop retrieval, affinity prediction, and structure-consistent loop generation.

Reference graph

Works this paper leans on

25 extracted references · 23 canonical work pages · cited by 1 Pith paper

[3]

In contrast, other baseline models are adversely affected by hallucinations stemming from conserved backbone structures

As we can observe, the performance of IgSeek improves with the inclusion of additional amino acids in the given structure, , which aligns with the fact that more input structural information can be encoded into the CDR representation. In contrast, other baseline models are adversely affected by hallucinations stemming from conserved backbone structures. N...

work page 2024
[4]

The neighborhood of a node vi, denoted as Ni, consists of the adjacency nodes of vi, that is, {vj|(vi, vj) ∈ E}

Each edge eij ∈ E denotes an interaction between vi and vj, if the Euclidean distance between their Cα atoms is within a threshold θ. The neighborhood of a node vi, denoted as Ni, consists of the adjacency nodes of vi, that is, {vj|(vi, vj) ∈ E}. CDR Sequence Design. Given the structure G = (E, V) of a CDR and the multi-channel 3D coordinate of each resid...

work page 2020
[6]

Besides, a dropout function (Srivastava et al.,

consists of a two-layer MLP with two Leaky Rectified Linear Unit (LeakyReLU) activation functions (Xu et al., 2015). Besides, a dropout function (Srivastava et al.,

work page 2015
[7]

is a deep learning–based method for protein sequence design that excels in both in silico and experimental evaluations. By leveraging a message-passing neural network with enhanced input features and edge updates, ProteinMPNN is capable of designing monomers, cyclic oligomers, protein nanoparticles, and protein-protein inter- faces, rescuing previously fa...

work page 2018
[8]

CDR Embedding Generation

For each node vi, ϕh aggregates the features of the adjacent edges intoh(l) aggi and combines the node representationh(l−1) i from the (l − 1)-th layer with the aggregated feature using a residual connection (He et al., 2016): h(l) aggi = X j∈Ni h(l) eij , (11) h(l) i = h(l−1) i + MLP CONCAT(h(l−1) i , h(l) aggi ) . CDR Embedding Generation. After the equ...

work page 2016
[9]

MEAN (Kong et al., 2023b) and dyMEAN (Kong et al., 2023a) employ graph neural networks to predict the structure and sequence of CDRs

utilize diffusion models to generate the structure and sequence of CDRs based on the framework regions and the target antigen, with DiffAb oriented for specific antigens. MEAN (Kong et al., 2023b) and dyMEAN (Kong et al., 2023a) employ graph neural networks to predict the structure and sequence of CDRs. Table 2 presents a comparative analysis of various a...

work page 2021
[12]

Hyperparameter Value Description Input noise ratio 0.15 Ratio of the input coordinates with added Gaussian noise

15 Table 3: Hyperparameters of IgSeek. Hyperparameter Value Description Input noise ratio 0.15 Ratio of the input coordinates with added Gaussian noise. noise scale 1 The standard deviation σ in the Gaussian noise. θ 10 ˚A The Euclidean distance threshold when constructing the graph G. MEGNN learning rate 5 × 10−3 Learning rate of MEGNN. weight decay 1 × ...

work page 2013
[13]

Notice that we remove CDR sequences that are identical to those in the dataset to elimi- nate redundancy in the dataset

before January 1, 2024, to sample our training set. Notice that we remove CDR sequences that are identical to those in the dataset to elimi- nate redundancy in the dataset. Following FoldSeek (Van Kempen et al., 2024), for each CDR in the SAbDab-before-2024 dataset, we randomly sample equal-length CDRs with TM-score large than 0.6 to generate training pai...

work page 2024

Show all 25 references

[15]

PyIgClassify cluster labels (North et al., 2011; Adolf-Bryfogle et al.,

Labels. PyIgClassify cluster labels (North et al., 2011; Adolf-Bryfogle et al.,

2011
[16]

are employed as ground-truth labels to assess the retrieval performance of antibody CDR regions. For each PDB structure containing an identified antibody heavy or light chain, PyIgClassify categorizes the con- formations of CDRs using a three-tier strategy: chain and position,...

2024
[18]

Official code is available at: https://github.com/steineggerlab/foldseek

represents tertiary amino acid interactions using 3D in- teraction (3Di) structural alphabet, achieving 4 to 5 orders of magnitude speed-up compared to traditional iterative or stochastic structure retrieval methods like CE (Shindyalov & Bourne, 1998), Dali (Holm, 2020), and T...

1998
[19]

is a deep learning–based method for protein sequence design that excels in both in silico and experimental evaluations, achieving a sequence recovery of 52.4% on native protein backbones, compared to 32.9% for Rosetta (Adolf-Bryfogle et al., 2018; Baek et al., 2021). By levera...

2018
[20]

Official code is available at: https://github.com/dauparas/ProteinMPNN

or AlphaFold (Jumper et al., 2021). Official code is available at: https://github.com/dauparas/ProteinMPNN. • ESM-IF1 (Hsu et al.,

2021
[22]

The profile of model weights is available at: https://zenodo.org/records/8164693

dataset for antibody design, outperforming generic protein models in se- quence recovery and structure robustness, especially for the hypervariable CDR-H3 loop. The profile of model weights is available at: https://zenodo.org/records/8164693. • AntiFold (Høie et al.,
[23]

on solved antibody structures from the SAbDab dataset (Dunbar et al., 2013; Schneider et al.,

2013
[24]

AntiFold excels in sequence recovery and structural similarity while also demonstrates stronger correlations in predicting antibody-antigen binding affinity in a zero-shot manner

and predicted antibody structures from the OAS dataset (Kovaltsuk et al., 2018; Olsen et al., 2022). AntiFold excels in sequence recovery and structural similarity while also demonstrates stronger correlations in predicting antibody-antigen binding affinity in a zero-shot mann...

2018
[1989]

Retrieval-augmented generation for large language models: A survey

Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Qianyu Guo, Meng Wang, and Haofen Wang. Retrieval-augmented generation for large language models: A survey. CoRR, abs/2312.10997, 2023a. Zhangyang Gao, Cheng Tan, Pablo Chac ´on, and Stan...

arXiv
[1998]

Dropout: a simple way to prevent neural networks from overfitting

Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting. JMLR, 15(1):1929–1958,

1929
[2007]

Se(3)-transformers: 3d roto- translation equivariant attention networks

Fabian Fuchs, Daniel Worrall, V olker Fischer, and Max Welling. Se(3)-transformers: 3d roto- translation equivariant attention networks. In NeurIPS, pp. 1970–1981,

1970
[2015]

Chengxin Zhang, Morgan Shine, Anna Marie Pyle, and Yang Zhang

URL https://arxiv.org/abs/1505.00853. Chengxin Zhang, Morgan Shine, Anna Marie Pyle, and Yang Zhang. Us-align: universal structure alignments of proteins, nucleic acids, and macromolecular complexes. Nature methods, 19(9): 1109–1115,

arXiv
[2018]

To evaluate the model efficiency, we utilize 5, 000 predicted CDR-H3 loops from the Observed Antibody Space (OAS) (Olsen et al., 2022), denoted as OAS-H3

to construct a test set with5, 111 receptors, referred to as STCRDab. To evaluate the model efficiency, we utilize 5, 000 predicted CDR-H3 loops from the Observed Antibody Space (OAS) (Olsen et al., 2022), denoted as OAS-H3. Redundant CDR loops are removed from the test set. S...

2022
[2021]

ESM-IF1 (Hsu et al.,

or AlphaFold (Jumper et al., 2021). ESM-IF1 (Hsu et al.,

2021
[2022]

The similarity of the search result is evaluated by US-align (Zhang et al., 2022)

by compressing data from ∼23 TB to ∼20 GB using vector embeddings, narrowing down candidates with a neural network. The similarity of the search result is evaluated by US-align (Zhang et al., 2022). Another state-of-the-art method, FoldSeek (Van Kempen et al., 2024), accelerat...

2022
[2023]

fine-tunes ProteinMPNN on the SAbDab (Dunbar et al., 2013; Schneider et al.,

2013
[2024]

org/abs/2405.03370

URL https://arxiv. org/abs/2405.03370. Liisa Holm. Using dali for protein structure comparison. Structural Bioinformatics: Methods and Protocols, pp. 29–42,

arXiv

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [3]

In contrast, other baseline models are adversely affected by hallucinations stemming from conserved backbone structures

As we can observe, the performance of IgSeek improves with the inclusion of additional amino acids in the given structure, , which aligns with the fact that more input structural information can be encoded into the CDR representation. In contrast, other baseline models are adversely affected by hallucinations stemming from conserved backbone structures. N...

work page 2024

[2] [4]

The neighborhood of a node vi, denoted as Ni, consists of the adjacency nodes of vi, that is, {vj|(vi, vj) ∈ E}

Each edge eij ∈ E denotes an interaction between vi and vj, if the Euclidean distance between their Cα atoms is within a threshold θ. The neighborhood of a node vi, denoted as Ni, consists of the adjacency nodes of vi, that is, {vj|(vi, vj) ∈ E}. CDR Sequence Design. Given the structure G = (E, V) of a CDR and the multi-channel 3D coordinate of each resid...

work page 2020

[3] [6]

Besides, a dropout function (Srivastava et al.,

consists of a two-layer MLP with two Leaky Rectified Linear Unit (LeakyReLU) activation functions (Xu et al., 2015). Besides, a dropout function (Srivastava et al.,

work page 2015

[4] [7]

is a deep learning–based method for protein sequence design that excels in both in silico and experimental evaluations. By leveraging a message-passing neural network with enhanced input features and edge updates, ProteinMPNN is capable of designing monomers, cyclic oligomers, protein nanoparticles, and protein-protein inter- faces, rescuing previously fa...

work page 2018

[5] [8]

CDR Embedding Generation

For each node vi, ϕh aggregates the features of the adjacent edges intoh(l) aggi and combines the node representationh(l−1) i from the (l − 1)-th layer with the aggregated feature using a residual connection (He et al., 2016): h(l) aggi = X j∈Ni h(l) eij , (11) h(l) i = h(l−1) i + MLP CONCAT(h(l−1) i , h(l) aggi ) . CDR Embedding Generation. After the equ...

work page 2016

[6] [9]

MEAN (Kong et al., 2023b) and dyMEAN (Kong et al., 2023a) employ graph neural networks to predict the structure and sequence of CDRs

utilize diffusion models to generate the structure and sequence of CDRs based on the framework regions and the target antigen, with DiffAb oriented for specific antigens. MEAN (Kong et al., 2023b) and dyMEAN (Kong et al., 2023a) employ graph neural networks to predict the structure and sequence of CDRs. Table 2 presents a comparative analysis of various a...

work page 2021

[7] [12]

Hyperparameter Value Description Input noise ratio 0.15 Ratio of the input coordinates with added Gaussian noise

15 Table 3: Hyperparameters of IgSeek. Hyperparameter Value Description Input noise ratio 0.15 Ratio of the input coordinates with added Gaussian noise. noise scale 1 The standard deviation σ in the Gaussian noise. θ 10 ˚A The Euclidean distance threshold when constructing the graph G. MEGNN learning rate 5 × 10−3 Learning rate of MEGNN. weight decay 1 × ...

work page 2013

[8] [13]

Notice that we remove CDR sequences that are identical to those in the dataset to elimi- nate redundancy in the dataset

before January 1, 2024, to sample our training set. Notice that we remove CDR sequences that are identical to those in the dataset to elimi- nate redundancy in the dataset. Following FoldSeek (Van Kempen et al., 2024), for each CDR in the SAbDab-before-2024 dataset, we randomly sample equal-length CDRs with TM-score large than 0.6 to generate training pai...

work page 2024