Pith. sign in

REVIEW 4 major objections 4 minor 28 references

HelixDesign-Binder: A Scalable Production-Grade Platform for Binder Design Built on HelixFold3

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A fully automated design platform is claimed to match or beat experimentally validated binders on predicted affinity for six targets.

desk verdict A useful engineering integration paper whose central quality claim is undermined by a selection artifact in the benchmark; the platform may be worth a look, but the evidence does not support the headline result. read the letter →

arxiv 2505.21873 v1 pith:IH3JXDCL submitted 2025-05-28 q-bio.BM cs.AIq-bio.QM

Jie Gao , Jun Li , Jing Hu , Shanzhuo Zhang , Kunrui Zhu , Yueyang Huang , Xiaonan Zhang , Xiaomin Fang This is my paper
classification q-bio.BMcs.AIq-bio.QM
keywords binderdesignHelixFold3inversefoldingipTMpredictedbindingfreeenergyhigh-throughputscreeningprotein-proteininteractionsequencediversity
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

HelixDesign-Binder is a fully automated platform that claims to make structure-based protein binder design practical at scale by wrapping the whole workflow—backbone generation, inverse folding, structure prediction, and multi-dimensional scoring—into one pipeline built around HelixFold3. On six previously characterized protein targets, the platform reports designed binders with interface confidence ipTM above 0.8, and predicted binding free energies that are more favorable than experimentally validated binders on five of six targets. It also reports that designed binders are highly sequence-diverse and share almost no sequence identity with known binders, and that larger sampling steadily improves predicted binding quality. The paper's point, if true, is that a user with a target sequence and a web browser can quickly obtain thousands of candidate binders worth experimental testing, without assembling a complicated computational stack. The evidence is entirely computational; no wet-lab binding measurements are presented.

What carries the argument

The load-bearing object is the integrated pipeline itself, and within it the scoring metrics that carry every quality claim. The pipeline extracts candidate binder backbones from structurally similar complexes in the structural database, filters them for diversity and plausibility, and then generates sequences with ESM-IF1, a structure-conditioned inverse-folding model adapted to multi-chain complexes. Each designed sequence is folded in complex with the target using HelixFold3, which outputs the interface predicted TM-score (ipTM), a confidence measure for the predicted binding interface, and candidates are ranked using FoldX-predicted binding free energy, PRODIGY contact and hydrophobicity statistics, and sequence-based fitness. The argument turns on ipTM and FoldX binding free energy as proxies for true binding: ipTM captures the geometric plausibility of the interface, the FoldX energy captures thermodynamic favorability, and the paper treats favorable values on both axes as jointly indicating a promising binder.

What would settle it

Take the top-ranked designed binders for one target such as VirB8, express and purify them, and measure binding to the target with a direct assay such as biolayer interferometry or surface plasmon resonance; if most do not bind or bind no better than the original validated binders, the platform's predictive claim fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a production-grade integration of existing structure-prediction and inverse-folding components can reliably turn a target sequence into diverse, structurally plausible, energetically favorable binder candidates. Using database-derived interface fragments as backbones, a multi-chain-adapted ESM-IF1 model proposes sequences, and HelixFold3 folds each candidate in complex with the target and scores the interface with ipTM. FoldX then estimates binding free energy, and additional physicochemical filters rank candidates. Across Interleukin-7 Receptor-α, VirB8, TrkA, InsR, FGFR2, and PDGFR, the designed binders are reported to match or exceed the ipTM of validated binders on four targets and to have lower predicted binding free energy on all except TrkA, with VirB8 designs reaching about -27.3 kcal/mol versus -12.4 kcal/mol for validated binders. The paper also observes a sampling scaling law: as the number of sampled sequences grows logarithmically, mean predicted binding free energy and top-100 ipTM improve across all six targets.

Load-bearing premise

The entire quality claim rests on the assumption that HelixFold3's ipTM scores and FoldX-predicted binding free energies reflect real binding affinity, even though the pipeline uses those same scores to select the candidates and the paper presents no wet-lab binding measurements.

Editorial extensions

If this is right

  • A target with a known structure can be submitted with only a sequence and a desired binder length range, and the platform will return thousands of ranked candidate binders.
  • Because designed binders show sequence identity at or below 0.1 against validated binders, the approach explores sequence space far from known solutions.
  • The reported sampling scaling law implies that access to large-scale computation directly translates into better predicted candidates, so users who can sample more should expect higher-quality top hits.
  • The joint use of ipTM, predicted binding free energy, and hydrophobic-contact counts gives a prioritization that no single score could provide, since the two main scores are only partially correlated.
  • Web access through the described interface is intended to let non-specialists run the full design cycle, lowering the practical barrier to binder design in academic and industrial settings.

Reading between the lines

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

  • Because the pipeline filters and evaluates with the same HelixFold3 ipTM scores, the reported quality may be partly an artifact of self-consistency; real binding-affinity measurements on top-ranked designs would be needed to confirm the benchmark.
  • The backbone-generation step depends on finding structurally similar complexes in structural databases, so targets lacking homologous complex structures may not get good starting backbones; extending the method to de novo backbone generation would be a natural next test.
  • If the sampling scaling law generalizes, small academic groups without high-performance computing access are at a systematic disadvantage, and publishing sample-size-to-quality curves for a standardized target set would let others calibrate how many candidates they need.
  • Because FoldX and ipTM are proxies with imperfect correlation to activity, top-ranked candidates are best treated as a focused shortlist for experimental screening rather than as confirmed binders.
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.

Referee Report

4 major / 4 minor

Summary. The paper presents HelixDesign-Binder, an integrated computational platform for protein binder design built on the authors' own structure prediction model HelixFold3. The pipeline combines backbone generation from PDB-derived fragments, ESM-IF-based sequence design, high-throughput HelixFold3 complex prediction, and multi-dimensional filtering and ranking using ipTM, predicted binding free energy (FoldX or PRODIGY), and physicochemical metrics. The authors benchmark the platform on six protein targets previously used in the Cao et al. (2022) binder design study, and report that designed binders achieve high ipTM scores, favorable FoldX binding free energies, high sequence diversity relative to validated binders, and a scaling relationship between sampling size and predicted binding quality. The central claim is that HelixDesign-Binder reliably produces diverse and high-quality binder candidates, some of which match or exceed experimentally validated designs in predicted binding affinity.

Significance. If the central claim were established, this would be a useful engineering contribution: the platform integrates several otherwise fragmented design steps, scales to thousands of candidates on cloud HPC resources, and is made accessible through a web interface. The authors also report a plausible scaling relationship between sampling effort and predicted binding quality. However, the evidence is entirely computational and suffers from a self-referential evaluation loop: candidates are filtered and ranked by HelixFold3 ipTM and then evaluated on that same HelixFold3 ipTM against experimentally validated binders. There are no wet-lab measurements, no statistical tests, no error bars, and no independent calibration of HelixFold3 on non-natural designed binders. As a result, the paper currently supports a claim about what the pipeline can generate in silico, but not the stronger quantitative claim that the designs 'match or exceed validated designs in predicted binding affinity' in a reliable or unbiased way.

major comments (4)
  1. [Sections 2.4 and 3.1, Figure 2(a)] The central benchmark is compromised by a selection artifact. Section 2.4 states that each candidate is 'filtered and ranked based on these sequence, structure, and physicochemical metrics,' with ipTM explicitly listed as a key structural metric. The designed-binder distributions shown in Figure 2(a) are therefore enriched for high ipTM by construction. Comparing this metric-enriched set against experimentally validated binders (which were selected by experimental activity, not by ipTM) on the same ipTM metric does not establish that the designs are better; it guarantees or inflates any apparent advantage. The authors should report ipTM distributions for the full unscreened candidate set, for a random subset, and for the final selected set, and should evaluate the final selection using metrics that were not part of the filtering criterion (for example, PRODIGY scores or, ideally, experimental measurements).
  2. [Sections 2.3 and 3.1] The evaluation is circular in an additional sense: HelixFold3 is the authors' own model, and it is used both to generate and filter candidate backbones/sequences and to compute the ipTM that serves as the primary success metric. The manuscript provides no independent calibration of HelixFold3's ipTM on designed, non-natural binders, nor any comparison against an independently developed structure prediction model. Without such calibration, the reported ipTM advantage could reflect systematic optimism of the in-house model for its own pipeline's outputs rather than genuine binder quality. The authors should provide calibration data on complexes with known binding affinity, or at least show agreement with an independent model such as AlphaFold-Multimer or AlphaFold3.
  3. [Sections 3.1 and 3.3] The quantitative claims about improved predicted binding free energy and the scaling law in Figure 2(d) are not supported by statistical evidence. The paper states that designed binders have more favorable FoldX energies than validated binders across all targets except TrkA, and that larger sampling improves average predicted binding free energy and mean ipTM, but no error bars, confidence intervals, sample sizes per target, or significance tests are reported. The number of designed binders per target (only stated as 'at least 1,000' passing the initial filter) is not given for the final benchmark distributions. The authors should report the full distributions, per-target sample sizes, and appropriate statistical comparisons (for example, permutation tests or bootstrap confidence intervals).
  4. [Section 3.2, Figure 2(e)] The selection of 'representative structures from the top-ranking region' is anecdotal and potentially cherry-picked. The text highlights that for three targets where the binding sites of designed and validated binders are 'spatially aligned,' the designed binders outperform the validated ones, but no quantitative definition of 'spatially aligned' is provided, and no criterion is given for why these three targets are singled out while other targets (e.g., TrkA and PDGFR) show worse or mixed results. This section should either present a systematic quantitative comparison over all targets and all designs, or be explicitly framed as illustrative case studies rather than evidence of general performance.
minor comments (4)
  1. [Section 3.1] Target names are inconsistent: 'INSULNR' appears where 'InsR' is used elsewhere, and 'TRKA' appears where 'TrkA' is used; Figure 3(a) labels the target 'IL1RA' while the text uses 'IL-7Rα'. Please standardize names throughout.
  2. [Figure 2(e)] The reported ΔGbind values in the structural visualization do not state their units (presumably kcal/mol) and are given without any uncertainty or context; please add units and state whether these are single representative designs or averages.
  3. [Figure 2(d)] It is not clear whether the curves in Figure 2(d) plot means over all sampled sequences, means over the top-100 binders, or some other quantity; the caption and text should specify this precisely, along with the number of designs used at each sampling size.
  4. [Section 3.1] The statements that designed binders 'consistently achieved ipTM scores above 0.8' and that for four targets they are 'comparable to or higher' than validated binders would benefit from precise numbers or at least a table, since the figure is not quantitatively legible from the text alone.

Circularity Check

1 steps flagged · score 6.0 of 10

The headline ipTM comparison is a selection artifact: Section 2.4 ranks candidates on HelixFold3 ipTM, then Section 3.1 presents the same ipTM as evidence of quality against experiment-selected binders.

  1. fitted input called prediction [Section 2.4 (Multi-dimensional Interaction Analysis) and Section 3.1 (Binder Design), Figures 2(a)-(b)]
    "Key metrics include the interface predicted TM-score (ipTM) ... Each candidate is filtered and ranked based on these sequence, structure, and physicochemical metrics. ... As shown in Figure 2(a) and (b), the designed binders consistently achieved ipTM scores above 0.8, indicating that the structure prediction model considered these binder–target complexes to be highly reliable."

    The ipTM values reported as evidence of binder quality are not independent measurements. Section 2.4 explicitly lists ipTM among the metrics used to filter and rank every candidate before final selection, and Section 3.1 then presents the surviving candidates' ipTM distribution as evidence that designs match or exceed experimentally validated binders. The validated binders were selected by experimental binding activity, not by ipTM, so the designed set is enriched for high ipTM by construction. Comparing a metric-enriched set against an experiment-selected set on that same metric guarantees or inflates the observed ipTM advantage.

full rationale

The central circular step is the use of HelixFold3 ipTM both as a selection filter (Section 2.4) and as the primary structural-quality benchmark against validated binders (Section 3.1). Because candidates are ranked and filtered on ipTM before the ipTM comparison is made, the reported 'designed binders achieve ipTM above 0.8' and the claim that designed binders match or exceed validated binders on this axis reduce by construction to the filter criterion. The FoldX-predicted binding free energy provides a partially independent signal, though it is correlated with the PRODIGY-derived affinity used in the same filtering step, and the paper's own caveat that 'ipTM and FoldX-predicted binding energy correlate with binding activity, they are not definitive predictors' further weakens the evaluative claim. No separate load-bearing self-citation chain was found beyond the same-team HelixFold3 citation, and the paper does benchmark against external validated binders under the same scoring scheme; however, the ipTM portion of the central claim is not self-contained. Score 6 reflects partial circularity: one of the two headline 'prediction' metrics is forced by the selection process, while the energy and diversity results retain some independent content.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper contributes a workflow integration, so its central claim depends on the validity of several adopted computational proxies and on the in-house HelixFold3 model. These are domain assumptions from prior work, and the paper offers no independent experimental evidence to ground them.

free parameters (2)
  • ipTM threshold (0.8) = 0.8
    The paper states 'designed binders consistently achieved ipTM scores above 0.8' and uses this as a quality bar. This cutoff is chosen post hoc from results, not derived from a theory, and it influences which designs are called high-quality.
  • Minimum sampled sequences per target (1,000) = 1,000
    Section 2.2 says at least 1,000 sequences pass the filter per target. This sampling budget is a design choice that affects the scaling results and computational cost, but it is not fitted to data.
assumptions (4)
  • domain assumption HelixFold3 predicts protein complexes with accuracy comparable to AlphaFold3
    Invoked in Sections 2.3 and 3.1 as the basis for ipTM-based evaluation; this is a claim from the authors' own prior technical report, not independently verified here.
  • domain assumption ipTM is positively correlated with binder binding capability
    Stated in Section 3.1 with citations to previous studies; the present paper provides no independent validation of this correlation for its designed sequences.
  • domain assumption FoldX predicted binding free energy is a valid proxy for binding affinity
    Used in Section 3.1 as the thermodynamic comparison metric; the paper cites prior evaluations but does not test it on its own designs.
  • domain assumption PDB fragment-based backbones offer sufficient coverage to find high-affinity binders
    The backbone generation in Section 2.1 relies on extracting interface fragments from structurally similar PDB complexes, assuming these fragments represent the space of viable binder geometries.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HelixDesign-Binder: A Scalable Production-Grade Platform for Binder Design Built on HelixFold3." pith.science (2026). https://pith.science/paper/IH3JXDCL

@misc{pith2026250521873,
  author       = {Pith},
  title        = {Pith review of: HelixDesign-Binder: A Scalable Production-Grade Platform for Binder Design Built on HelixFold3},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IH3JXDCL}},
  note         = {Machine review of arXiv:2505.21873}
}
read the original abstract

Protein binder design is central to therapeutics, diagnostics, and synthetic biology, yet practical deployment remains challenging due to fragmented workflows, high computational costs, and complex tool integration. We present HelixDesign-Binder, a production-grade, high-throughput platform built on HelixFold3 that automates the full binder design pipeline, from backbone generation and sequence design to structural evaluation and multi-dimensional scoring. By unifying these stages into a scalable and user-friendly system, HelixDesign-Binder enables efficient exploration of binder candidates with favorable structural, energetic, and physicochemical properties. The platform leverages Baidu Cloud's high-performance infrastructure to support large-scale design and incorporates advanced scoring metrics, including ipTM, predicted binding free energy, and interface hydrophobicity. Benchmarking across six protein targets demonstrates that HelixDesign-Binder reliably produces diverse and high-quality binders, some of which match or exceed validated designs in predicted binding affinity. HelixDesign-Binder is accessible via an interactive web interface in PaddleHelix platform, supporting both academic research and industrial applications in antibody and protein binder development.

Figures

Figures reproduced from arXiv: 2505.21873 by the authors.

Figure 1
Figure 1. HelixDesign-Binder for structure-based binder design. HelixDesign-Binder begins with a user-defined target [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Results of HelixDesign-Binder on six protein targets. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Correlation between predicted binding free energies (kcal/mol) and ipTM score for six protein targets: (a) [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Example input interface of the HelixDesign-Binder Server. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Example output display page of the HelixDesign-Binder Server. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references · 19 canonical work pages

  1. [1]

    Protein complex prediction with alphafold-multimer.biorxiv, pages 2021–10, 2021

    Richard Evans, Michael O’Neill, Alexander Pritzel, Natasha Antropova, Andrew Senior, Tim Green, Augustin Žídek, Russ Bates, Sam Blackwell, Jason Yim, et al. Protein complex prediction with alphafold-multimer.biorxiv, pages 2021–10, 2021

    work page 2021

  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]

    Technical report of helixfold3 for biomolecular structure prediction.arXiv preprint arXiv:2408.16975, 2024

    Lihang Liu, Shanzhuo Zhang, Yang Xue, Xianbin Ye, Kunrui Zhu, Yuxin Li, Yang Liu, Jie Gao, Wenlai Zhao, Hongkun Yu, et al. Technical report of helixfold3 for biomolecular structure prediction.arXiv preprint arXiv:2408.16975, 2024

    arXiv 2024

  4. [4]

    De novo design of high-affinity protein binders with alphaproteo.arXiv preprint arXiv:2409.08022, 2024

    Vinicius Zambaldi, David La, Alexander E Chu, Harshnira Patani, Amy E Danson, Tristan OC Kwan, Thomas Frerix, Rosalia G Schneider, David Saxton, Ashok Thillaisundaram, et al. De novo design of high-affinity protein binders with alphaproteo.arXiv preprint arXiv:2409.08022, 2024

    arXiv 2024

  5. [5]

    Bindcraft: one-shot design of functional protein binders.bioRxiv, pages 2024–09, 2024

    Martin Pacesa, Lennart Nickel, Christian Schellhaas, Joseph Schmidt, Ekaterina Pyatova, Lucas Kissling, Patrick Barendse, Jagrity Choudhury, Srajan Kapoor, Ana Alcaraz-Serna, et al. Bindcraft: one-shot design of functional protein binders.bioRxiv, pages 2024–09, 2024

    work page 2024

  6. [6]

    De novo design of protein structure and function with rfdiffusion.Nature, 620(7976):1089–1100, 2023

    Joseph L Watson, David Juergens, Nathaniel R Bennett, Brian L Trippe, Jason Yim, Helen E Eisenach, Woody Ahern, Andrew J Borst, Robert J Ragotte, Lukas F Milles, et al. De novo design of protein structure and function with rfdiffusion.Nature, 620(7976):1089–1100, 2023

    2023

  7. [7]

    Atom level enzyme active site scaffolding using rfdiffusion2

    Woody Ahern, Jason Yim, Doug Tischer, Saman Salike, Seth Woodbury, Donghyo Kim, Indrek Kalvet, Yakov Kipnis, Brian Coventry, Han Altae-Tran, et al. Atom level enzyme active site scaffolding using rfdiffusion2. bioRxiv, pages 2025–04, 2025

    work page 2025

  8. [8]

    Learning inverse folding from millions of predicted structures

    Chloe Hsu, Robert Verkuil, Jason Liu, Zeming Lin, Brian Hie, Tom Sercu, Adam Lerer, and Alexander Rives. Learning inverse folding from millions of predicted structures. InInternational conference on machine learning, pages 8946–8970. PMLR, 2022

    2022

Show all 28 references
  1. [9]

    Robust deep learning–based protein sequence design using proteinmpnn.Science, 378(6615):49–56, 2022

    Justas Dauparas, Ivan Anishchenko, Nathaniel Bennett, Hua Bai, Robert J Ragotte, Lukas F Milles, Basile IM Wicky, Alexis Courbet, Rob J de Haas, Neville Bethel, et al. Robust deep learning–based protein sequence design using proteinmpnn.Science, 378(6615):49–56, 2022

    2022

  2. [10]

    The foldx web server: an online force field.Nucleic acids research, 33(suppl_2):W382–W388, 2005

    Joost Schymkowitz, Jesper Borg, Francois Stricher, Robby Nys, Frederic Rousseau, and Luis Serrano. The foldx web server: an online force field.Nucleic acids research, 33(suppl_2):W382–W388, 2005

    2005

  3. [11]

    Prodigy: a web server for predicting the binding affinity of protein–protein complexes.Bioinformatics, 32(23):3676–3678, 2016

    Li C Xue, João Pglm Rodrigues, Panagiotis L Kastritis, Alexandre Mjj Bonvin, and Anna Vangone. Prodigy: a web server for predicting the binding affinity of protein–protein complexes.Bioinformatics, 32(23):3676–3678, 2016

    2016

  4. [12]

    Improving de novo protein binder design with deep learning.Nature Communications, 14(1):2625, 2023

    Nathaniel R Bennett, Brian Coventry, Inna Goreshnik, Buwei Huang, Aza Allen, Dionne Vafeados, Ying Po Peng, Justas Dauparas, Minkyung Baek, Lance Stewart, et al. Improving de novo protein binder design with deep learning.Nature Communications, 14(1):2625, 2023

    2023

  5. [13]

    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

  6. [14]

    Design of protein-binding proteins from the target structure alone.Nature, 605(7910):551–560, 2022

    Longxing Cao, Brian Coventry, Inna Goreshnik, Buwei Huang, William Sheffler, Joon Sung Park, Kevin M Jude, Iva Markovi´c, Rameshwar U Kadam, Koen HG Verschueren, et al. Design of protein-binding proteins from the target structure alone.Nature, 605(7910):551–560, 2022

    2022

  7. [15]

    Protein data bank (pdb): the single global macromolecular structure archive.Protein crystallography: methods and protocols, pages 627–641, 2017

    Stephen K Burley, Helen M Berman, Gerard J Kleywegt, John L Markley, Haruki Nakamura, and Sameer Velankar. Protein data bank (pdb): the single global macromolecular structure archive.Protein crystallography: methods and protocols, pages 627–641, 2017

    2017

  8. [16]

    Evaluation of alphafold 3’s protein–protein complexes for predicting binding free energy changes upon mutation.Journal of Chemical Information and Modeling, 64(16):6676–6683, 2024

    JunJie Wee and Guo-Wei Wei. Evaluation of alphafold 3’s protein–protein complexes for predicting binding free energy changes upon mutation.Journal of Chemical Information and Modeling, 64(16):6676–6683, 2024

    2024

  9. [17]

    Implementing and assessing an alchemical method for calculating protein–protein binding free energy.Journal of chemical theory and computation, 17(4):2457–2464, 2021

    Dharmeshkumar Patel, Jagdish Suresh Patel, and F Marty Ytreberg. Implementing and assessing an alchemical method for calculating protein–protein binding free energy.Journal of chemical theory and computation, 17(4):2457–2464, 2021

    2021

  10. [18]

    Assessment of software methods for estimating protein-protein relative binding affinities.PLoS One, 15(12):e0240573, 2020

    Tawny R Gonzalez, Kyle P Martin, Jonathan E Barnes, Jagdish Suresh Patel, and F Marty Ytreberg. Assessment of software methods for estimating protein-protein relative binding affinities.PLoS One, 15(12):e0240573, 2020. 9 HelixDesign-Binder

    2020

  11. [19]

    Ab-bind: antibody binding mutational database for computational affinity predictions.Protein Science, 25(2):393–409, 2016

    Sarah Sirin, James R Apgar, Eric M Bennett, and Amy E Keating. Ab-bind: antibody binding mutational database for computational affinity predictions.Protein Science, 25(2):393–409, 2016

    2016

  12. [20]

    Structural and biophysical studies of the human il-7/il-7rα complex.Structure, 17(1):54–65, 2009

    Craig A McElroy, Julie A Dohm, and Scott TR Walsh. Structural and biophysical studies of the human il-7/il-7rα complex.Structure, 17(1):54–65, 2009

    2009

  13. [21]

    Structural insight into how bacteria prevent interference between multiple divergent type iv secretion systems.MBio, 6(6):10–1128, 2015

    Joseph J Gillespie, Isabelle QH Phan, Holger Scheib, Sandhya Subramanian, Thomas E Edwards, Stephanie S Lehman, Hanna Piitulainen, M Sayeedur Rahman, Kristen E Rennoll-Bankert, Bart L Staker, et al. Structural insight into how bacteria prevent interference between multiple div...

    2015

  14. [22]

    Crystal structure of nerve growth factor in complex with the ligand-binding domain of the trka receptor.Nature, 401(6749):184–188, 1999

    Christian Wiesmann, Mark H Ultsch, Steven H Bass, and Abraham M de V os. Crystal structure of nerve growth factor in complex with the ligand-binding domain of the trka receptor.Nature, 401(6749):184–188, 1999

    1999

  15. [23]

    Higher-resolution structure of the human insulin receptor ectodomain: multi-modal inclusion of the insert domain.Structure, 24(3):469–476, 2016

    Tristan I Croll, Brian J Smith, Mai B Margetts, Jonathan Whittaker, Michael A Weiss, Colin W Ward, and Michael C Lawrence. Higher-resolution structure of the human insulin receptor ectodomain: multi-modal inclusion of the insert domain.Structure, 24(3):469–476, 2016

    2016

  16. [24]

    Crystal structures of two fgf-fgfr complexes reveal the determinants of ligand-receptor specificity.Cell, 101(4):413–424, 2000

    Alexander N Plotnikov, Stevan R Hubbard, Joseph Schlessinger, and Moosa Mohammadi. Crystal structures of two fgf-fgfr complexes reveal the determinants of ligand-receptor specificity.Cell, 101(4):413–424, 2000

    2000

  17. [25]

    Structures of a platelet-derived growth factor/propeptide complex and a platelet-derived growth factor/receptor complex

    Ann Hye-Ryong Shim, Heli Liu, Pamela J Focia, Xiaoyan Chen, P Charles Lin, and Xiaolin He. Structures of a platelet-derived growth factor/propeptide complex and a platelet-derived growth factor/receptor complex. Proceedings of the National Academy of Sciences, 107(25):11307–11...

    2010

  18. [26]

    Improved prediction of protein-protein interactions using alphafold2.Nature communications, 13(1):1265, 2022

    Patrick Bryant, Gabriele Pozzati, and Arne Elofsson. Improved prediction of protein-protein interactions using alphafold2.Nature communications, 13(1):1265, 2022

    2022

  19. [27]

    Enhanced protein-protein interaction discovery via alphafold-multimer.bioRxiv, pages 2024–02, 2024

    Ah-Ram Kim, Yanhui Hu, Aram Comjean, Jonathan Rodiger, Stephanie E Mohr, and Norbert Perrimon. Enhanced protein-protein interaction discovery via alphafold-multimer.bioRxiv, pages 2024–02, 2024

    2024

  20. [28]

    Protein modeling and structure-based drug design

    Gerhard Klebe. Protein modeling and structure-based drug design. InDrug Design: From Structure and Mode-of-Action to Rational Design Concepts, pages 309–321. Springer, 2025. 10 HelixDesign-Binder Figure 4: Example input interface of the HelixDesign-Binder Server. 11 HelixDesig...

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.