Pith. sign in

REVIEW 5 major objections 4 minor 1 cited by

Resilient Biosecurity in the Era of AI-Enabled Bioweapons

T0 review · 5 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper argues that PPI-based inference filters, benchmarked on known virus–host pairs and four confirmed SARS-CoV-2 mutants, fail too often to serve as reliable biosecurity screens, and that policy should switch from prediction to rapid

desk verdict A small, concrete benchmark shows PPI predictors miss many viral-host interactions and all tested mutants, but the abstract's broad 'filters are inadequate' claim is threshold-dependent and overreaches the data. read the letter →

arxiv 2509.02610 v1 pith:KXPNYUDG submitted 2025-08-30 q-bio.QM cs.AI

Jonathan Feldman , Tal Feldman This is my paper
classification q-bio.QMcs.AI
keywords biosecurityprotein–proteininteractionpredictioninference-timefilteringgenerativeproteindesignSARS-CoV-2virus–hostinteractionsAIbioweaponsresilience
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

This paper tries to establish that the standard technical defense against AI-designed biological weapons—inference-time filters that predict whether a viral protein binds a human receptor—cannot be trusted, even for well-known viruses. Benchmarking three leading protein–protein interaction predictors on thirteen documented virus–host binding pairs, the authors find that each tool misses between roughly 30% and 50% of known interactions, and all three fail to flag any of the four experimentally confirmed SARS-CoV-2 spike mutants. They conclude that prediction-based screening is structurally mismatched to synthetic threats and that biosecurity should pivot from containment to response: fast experimental validation, flexible manufacturing, and faster regulatory pathways. If this is right, the near-term policy question is not how to filter model outputs but how to prepare to respond after a threat emerges.

What carries the argument

The load-bearing object is the PPI prediction pipeline made of three tools—AlphaFold 3, AF3Complex, and SpatialPPIv2—each reduced to a binary 'interaction / no interaction' call via a fixed confidence threshold: ipTM ≥ 0.6 for AlphaFold 3, pIS ≥ 0.38 for AF3Complex, and probability ≥ 0.5 for SpatialPPIv2. The argument's success/failure labels in Tables 1 and 2 depend on these cutoffs; the paper's inference is that because the models miss known and experimentally confirmed interactions at these operating points, they cannot serve as reliable filters for unknown threats.

What would settle it

Run a threshold-free benchmark on the same thirteen complexes and four mutants—using ranking or AUROC instead of fixed cutoffs, and optionally recalibrating thresholds on a held-out set of known viral–host interactions. If a recalibrated threshold or a ranking recovers most known interactions and separates the high-affinity mutants from non-binders, then the paper's conclusion that the models themselves are inadequate for viral–host filtering would be substantially weakened.

Watch

Extended reading notes

Core claim

The paper's central claim is that current protein–protein interaction (PPI) prediction models, used as inference-time biosecurity filters, fail to generalize to viral–host interactions even when those interactions are well characterized and were present in training data. In the authors' benchmark, AlphaFold 3, AF3Complex, and SpatialPPIv2 were given thirteen known virus–receptor pairs; AlphaFold 3 missed about half, SpatialPPIv2 about 40%, and AF3Complex about 30%, including the SARS-CoV-2 spike–ACE2 interaction central to COVID-19. In a further test, none of the three tools predicted a meaningful interaction for any of four experimentally validated SARS-CoV-2 spike mutants with confirmed AC

Load-bearing premise

The paper's failure counts assume the standard confidence cutoffs (0.6, 0.38, 0.5) honestly reflect whether a viral–host interaction is real; if those cutoffs are miscalibrated for viruses, the benchmark may be measuring threshold error rather than model failure.

Editorial extensions

If this is right

  • A biosecurity framework that relies on PPI prediction as a screening gate will tolerate false negatives on high-consequence inputs; the benchmark shows this is not hypothetical.
  • The four-mutant result implies that even affinity-enhancing mutations can be invisible to current predictors, so sequence-similarity screening alone cannot compensate.
  • Resources should be redirected from improving filters toward experimental validation capacity, high-throughput screening, and manufacturing readiness.
  • Regulatory pathways for AI-generated therapeutics should be designed for emergency speed, with pre-authorized fast-track mechanisms and post-deployment monitoring.

Reading between the lines

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

  • The benchmark's fixed thresholds conflate model inadequacy with threshold miscalibration; a threshold-free evaluation (ranking or AUROC) on the same thirteen complexes would clarify whether the signal is absent or merely mis-calibrated.
  • The erratic confidence shifts on mutants suggest the models' scores are not monotone in binding affinity; this can be tested directly against large deep-mutational scanning datasets, turning the mutant case study into a quantitative correlation test.
  • If resilience becomes the goal, a natural success metric is time from sequence release to validated countermeasure; comparing that metric across existing public-health exercises would make the policy argument empirically testable.
  • The authors' policy conclusion—shift to response—does not depend on PPI tools being forever inadequate; it could survive even large model improvements as long as experimental ground truth remains the bottleneck.
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

5 major / 4 minor

Summary. The paper evaluates three protein–protein interaction (PPI) prediction tools — SpatialPPIv2, AlphaFold 3, and AF3Complex — on thirteen known viral–host interactions (Table 1) and on four experimentally characterized SARS-CoV-2 spike RBD–ACE2 mutants (Table 2). It reports that, at fixed confidence thresholds, the tools miss a substantial fraction of known interactions and fail to call any of the four mutants as positive. From this, the paper argues that current inference-time biosecurity filters are inadequate even for known threats and are unlikely to detect novel engineered proteins, and it proposes a policy shift toward response-oriented infrastructure: rapid experimental validation, biomanufacturing capacity, and regulatory reform. The paper is interdisciplinary, combining a small benchmark study with a broader biosecurity-policy argument.

Significance. If the empirical claim were established with proper calibration and statistical support, the finding would be policy-relevant: it would challenge the assumption that computational PPI filters can serve as dependable biosecurity screening tools. The study uses external ground truth (PDB structures and published binding-affinity measurements), so the central negative result is not circular. The raw confidence scores in Table 2 are transparently reported, and the selection of viruses spans several families. However, the empirical basis is small (n=13 interactions, n=4 mutants), the success/failure labels depend entirely on unvalidated thresholds, and no negative controls or uncertainty estimates are provided. The manuscript's headline conclusion is therefore broader than the evidence currently supports.

major comments (5)
  1. [§2, Tables 1–2] All success/failure calls are made by applying fixed thresholds (SpatialPPIv2 ≥0.5, AlphaFold3 ipTM ≥0.6, AF3Complex pIS ≥0.38) to raw confidence scores. The manuscript provides no evidence that these cutoffs are valid operating points for viral–host interactions, and no comparison against non-interacting decoy pairs or threshold-free discriminative analysis (e.g., AUC/rank-based metrics). A model could rank true interactors above non-interactors while still falling below a miscalibrated cutoff. Thus the central 'filters are inadequate' claim, and specifically the all-fail mutant result in Table 2, is threshold-dependent and does not yet reject the null that these models are usable at a properly calibrated operating point. The authors should either calibrate thresholds for viral–host complexes or report discrimination performance over positives and negatives.
  2. [§2, Tables 1–2] The quantitative support is statistically thin: n=13 interactions and n=4 mutants, with no confidence intervals, significance tests, or error bars. The statement that SpatialPPIv2, AlphaFold3, and AF3Complex 'misidentified approximately 40%, 50%, and 30%' respectively is a point estimate subject to large uncertainty (with n=13, a 95% Wilson interval for 40% spans roughly 18–67%). The mutant experiment has only four points, two of which are affinity-reducing; drawing the strong conclusion that the models are 'insensitive to combinatorial mutational effects' from 4/4 failures below an uncalibrated threshold is not robust. The authors should add statistical uncertainty and, ideally, a larger mutant panel or at least threshold-free analysis of the raw scores.
  3. [Abstract and §4] The abstract and conclusion generalize to 'current predictive filters' and 'inference-time filtering', but the benchmark tests only three PPI predictors. Section 2 explicitly concedes that sequence similarity searches are effective for identifying known pathogenic proteins and variants. Since the abstract's strongest claim — that current filters are inadequate for reliably flagging even known threats — is framed to include sequence alignment, the benchmark does not support that generalization. Either the scope should be narrowed to PPI-based filters, or sequence screening tools should be benchmarked under the same protocol.
  4. [§5] The data and model availability statements are not actionable: they refer to 'this GitHub repository' and 'this web server' without URLs, and the exact input sequences, model versions, parameters, and commands are not given. For AlphaFold 3, the server version and any non-default settings are unspecified. This prevents independent verification of the benchmark, which is load-bearing for the paper's central claim. The authors should provide full input/output data, code, and versioned links.
  5. [§2, paragraph on training data] The paper asserts that 'many of the PPIs included in the benchmarking study were present in the training datasets of all three models', citing general model papers. No specific evidence is provided that the particular PDB structures used here (e.g., 6M0J, 1GC1) were in the training sets. Since the 'trained on many of the same viruses' point is used to make the failures more damning, this claim needs verification (e.g., training-set membership analysis or explicit documentation from the model releases).
minor comments (4)
  1. [Table 2 and surrounding text] The text states that 'for both beneficial and detrimental mutations, SpatialPPIv2 and AlphaFold3 increased their predictive confidence', but Table 2 shows SpatialPPIv2 confidence change of +0.00 for the +2.14 mutant, which is no increase. The statement should be corrected to match the table.
  2. [Throughout] The virus name is written inconsistently: 'SARS-CoV-2' in the text and Tables 1, but 'Sars-Cov-2' in the Table 2 caption. Please standardize.
  3. [§2, Table 2 caption] The caption defines Δ log10 KD such that negative indicates increased affinity, but this is the opposite of the usual KD convention (lower KD = higher affinity). The sign convention is internally consistent with rows, but it will confuse readers; please clarify by stating Kd directly.
  4. [§5] The statement 'The SpatialPPIv2 model used for the analyses in this manuscript can be accessed at this GitHub repository' is incomplete: the URL is missing. The same applies to the AF3Complex repository and the AlphaFold 3 web server link.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the benchmark applies external ground-truth interactions and externally defined confidence thresholds; no fitted parameter is renamed as a prediction.

full rationale

The paper's central claim—that current PPI filters fail to flag known viral–host interactions—rests on a benchmark in which the inputs are (a) thirteen well-characterized viral–host interaction pairs with PDB structures, (b) four experimentally characterized SARS-CoV-2 spike mutants from an external study, and (c) confidence-score thresholds (SpatialPPIv2 ≥ 0.5, AlphaFold3 ipTM ≥ 0.6, AF3Complex pIS ≥ 0.38) taken from prior work, including an author's own AF3Complex preprint for pIS. None of these thresholds are fitted in the present paper to the test cases, and the ground-truth labels are not derived from the model outputs. The AF3Complex self-citation provides a parameter-free operating point defined in earlier work, not an assumption that includes the present target results, so it is not load-bearing circularity. Concerns about threshold calibration or over-generalization from PPI tools to all 'predictive filters' are correctness or evidentiary issues, not circular reasoning. The derivation chain is therefore self-contained with respect to circularity: the failures reported are genuine empirical outcomes under stated, externally sourced decision rules.

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

The central claim depends on three externally chosen thresholds, which are not fitted here but are load-bearing for all binary outcomes. The ground truth and representativeness assumptions are also domain assumptions that could change the results if violated. No new entities are introduced.

free parameters (3)
  • AlphaFold3 ipTM threshold = 0.6
    Adopted from prior studies (Wee and Wei 2024) as criterion for meaningful PPI; not fitted in this paper, but binary results in Tables 1 and 2 depend on it.
  • AF3Complex pIS threshold = 0.38
    Adopted from previous AF3Complex work; not fitted here; used as threshold for PPI.
  • SpatialPPIv2 probability threshold = 0.5
    Standard default for SpatialPPIv2; not fitted here; used to label predictions.
assumptions (4)
  • domain assumption PDB structures (e.g., 6M0J, 4L72) represent validated viral-host interactions
    Section 5 lists PDB IDs as ground truth for the benchmark; if any are wrong or not true in vivo interactions, the success/failure labels shift.
  • domain assumption PPI prediction is a meaningful proxy for pathogenicity in biosecurity filtering
    Section 2 motivates PPI prediction as a safety filter; the entire policy conclusion assumes that missing a PPI equals missing a threat.
  • domain assumption The 13 selected viral-host pairs are representative of biosecurity-relevant interactions
    Section 2 selects interactions from PDB with resolved structures; results may not generalize to low-similarity or engineered proteins.
  • domain assumption The four SARS-CoV-2 mutants from Moulana et al. are representative of computationally designed variants
    Section 2 uses four mutants with measured affinity changes; they may not represent the distribution of AI-designed proteins.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Resilient Biosecurity in the Era of AI-Enabled Bioweapons." pith.science (2026). https://pith.science/paper/KXPNYUDG

@misc{pith2026250902610,
  author       = {Pith},
  title        = {Pith review of: Resilient Biosecurity in the Era of AI-Enabled Bioweapons},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KXPNYUDG}},
  note         = {Machine review of arXiv:2509.02610}
}
read the original abstract

Recent advances in generative biology have enabled the design of novel proteins, creating significant opportunities for drug discovery while also introducing new risks, including the potential development of synthetic bioweapons. Existing biosafety measures primarily rely on inference-time filters such as sequence alignment and protein-protein interaction (PPI) prediction to detect dangerous outputs. In this study, we evaluate the performance of three leading PPI prediction tools: AlphaFold 3, AF3Complex, and SpatialPPIv2. These models were tested on well-characterized viral-host interactions, such as those involving Hepatitis B and SARS-CoV-2. Despite being trained on many of the same viruses, the models fail to detect a substantial number of known interactions. Strikingly, none of the tools successfully identify any of the four experimentally validated SARS-CoV-2 mutants with confirmed binding. These findings suggest that current predictive filters are inadequate for reliably flagging even known biological threats and are even more unlikely to detect novel ones. We argue for a shift toward response-oriented infrastructure, including rapid experimental validation, adaptable biomanufacturing, and regulatory frameworks capable of operating at the speed of AI-driven developments.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Harmonizing AI Safety Thresholds

    cs.AI 2026-07 conditional novelty 5.0 of 10

    The authors propose harmonized AI capability floors: non-zero full-chain TLO cyber completion triggers safeguards, and AI progress at 5× trend for 3 months triggers safeguards, with biorisk left as a diagnostic.

Reference graph

Works this paper leans on

57 extracted references · 43 canonical work pages · cited by 1 Pith paper

  1. [1]

    Ruffolo and Ali Madani

    Jeffrey A. Ruffolo and Ali Madani. Designing proteins with language models. Nature Biotechnology, 42(2): 200–202, February 2024. ISSN 1546-1696. doi: 10.1038/s41587-024-02123-4. URL https://doi.org/10. 1038/s41587-024-02123-4

    work page doi:10.1038/s41587-024-02123-4 2024

  2. [2]

    Learning protein fitness landscapes with deep mutational scanning data from multiple sources

    Lin Chen, Zehong Zhang, Zhenghao Li, Rui Li, Ruifeng Huo, Lifan Chen, Dingyan Wang, Xiaomin Luo, Kaixian Chen, Cangsong Liao, and Mingyue Zheng. Learning protein fitness landscapes with deep mutational scanning data from multiple sources. Cell Systems, 14(8):706–721.e5, 2023. ISSN 2405-4712. doi: https://doi.org/10.1016/j. cels.2023.07.003. URL https://ww...

    doi:10.1016/j 2023

  3. [3]

    Protein large language models: A comprehensive survey

    Yijia Xiao, Wanjia Zhao, Junkai Zhang, Yiqiao Jin, Han Zhang, Zhicheng Ren, Renliang Sun, Haixin Wang, Guancheng Wan, Pan Lu, Xiao Luo, Yu Zhang, James Zou, Yizhou Sun, and Wei Wang. Protein large language models: A comprehensive survey. 2025. URL https://arxiv.org/abs/2502.17504

    arXiv 2025

  4. [4]

    Language models enable zero-shot prediction of the effects of mutations on protein function

    Joshua Meier, Roshan Rao, Robert Verkuil, Jason Liu, Tom Sercu, and Alex Rives. Language models enable zero-shot prediction of the effects of mutations on protein function. In M. Ranzato, A. Beygelzimer, Y . Dauphin, P.S. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Processing Systems, volume 34, pages 29287–29303. Curran Associat...

    work page 2021

  5. [5]

    Hie, Kevin K

    Brian L. Hie, Kevin K. Yang, and Peter S. Kim. Evolutionary velocity with protein language models predicts evolutionary dynamics of diverse proteins. Cell Systems, 13(4):274–285.e6, April 2022. ISSN 2405-4720. doi: 10.1016/j.cels.2022.01.003

    doi:10.1016/j.cels.2022.01.003 2022

  6. [6]

    Watson, David Juergens, Nathaniel R

    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, Basile I. M. Wicky, Nikita Hanikel, Samuel J. Pellock, Alexis Courbet, William Sheffler, Jue Wang, Preetham Venkatesh, Isaac Sappington, Susana Vázquez Torres, Anna Lauko, Valentin De Bort...

    2023

  7. [7]

    Applying computational protein design to therapeutic antibody discovery -- current state and perspectives

    Weronika Bielska, Igor Jaszczyszyn, Pawel Dudzic, Bartosz Janusz, Dawid Chomicz, Sonia Wrobel, Victor Greiff, Ryan Feehan, Jared Adolf-Bryfogle, and Konrad Krawczyk. Applying computational protein design to therapeutic antibody discovery – current state and perspectives, 2025. URL https://arxiv.org/abs/2503.00913. 7 Resilient Biosecurity

    work page Pith review arXiv 2025

  8. [8]

    Without Safeguards, AI-Biology Integration Risks Accelerating Future Pandemics

    Dianzhuo Wang, Marian Huot, Zechen Zhang, Kaiyi Jiang, Eugene I Shakhnovich, and Kevin M Esvelt. Without Safeguards, AI-Biology Integration Risks Accelerating Future Pandemics. 2025. doi: 10.13140/RG.2.2.29765. 15849. URL https://rgdoi.net/10.13140/RG.2.2.29765.15849. Publisher: Unpublished

    work page arXiv 2025

Show all 57 references
  1. [9]

    Few-Shot Viral Variant Detection via Bayesian Active Learning and Biophysics, March 2025

    Marian Huot, Dianzhuo Wang, Jiacheng Liu, and Eugene Shakhnovich. Few-Shot Viral Variant Detection via Bayesian Active Learning and Biophysics, March 2025. URL https://www.biorxiv.org/content/10. 1101/2025.03.12.642881v1. Pages: 2025.03.12.642881 Section: New Results

    2025

  2. [10]

    Brock, Javier A

    Noor Youssef, Sarah Gurev, Fadi Ghantous, Kelly P. Brock, Javier A. Jaimes, Nicole N. Thadani, Ann Dauphin, Amy C. Sherman, Leonid Yurkovetskiy, Daria Soto, Ralph Estanboulieh, Ben Kotzen, Pascal Notin, Aaron W. Kollasch, Alexander A. Cohen, Sandra E. Dross, Jesse Erasmus, Deb...

    2025

  3. [11]

    Zero-shot prediction of mutation effects with multimodal deep representation learning guides protein engineering

    Peng Cheng, Cong Mao, Jin Tang, Sen Yang, Yu Cheng, Wuke Wang, Qiuxi Gu, Wei Han, Hao Chen, Sihan Li, Yaofeng Chen, Jianglin Zhou, Wuju Li, Aimin Pan, Suwen Zhao, Xingxu Huang, Shiqiang Zhu, Jun Zhang, Wenjie Shu, and Shengqi Wang. Zero-shot prediction of mutation effects with...

    2024

  4. [12]

    Sofroniew, Deniz Oktay, Zeming Lin, Robert Verkuil, Vincent Q

    Thomas Hayes, Roshan Rao, Halil Akin, Nicholas J. Sofroniew, Deniz Oktay, Zeming Lin, Robert Verkuil, Vincent Q. Tran, Jonathan Deaton, Marius Wiggert, Rohil Badkundri, Irhum Shafkat, Jun Gong, Alexander Derry, Raul S. Molina, Neil Thomas, Yousuf Khan, Chetan Mishra, Carolyn K...

    2024

  5. [13]

    A call for built-in biosecurity safeguards for generative AI tools

    Mengdi Wang, Zaixi Zhang, Amrit Singh Bedi, Alvaro Velasquez, Stephanie Guerra, Sheng Lin-Gibson, Le Cong, Yuanhao Qu, Souradip Chakraborty, Megan Blewett, Jian Ma, Eric Xing, and George Church. A call for built-in biosecurity safeguards for generative AI tools. Nature Biotech...

    2025 doi

  6. [14]

    Protein design meets biosecurity

    David Baker and George Church. Protein design meets biosecurity. Science, 383(6681):349–349, 2024. doi: 10.1126/science.ado1671. URL https://www.science.org/doi/abs/10.1126/science.ado1671

    2024 doi

  7. [15]

    Constance J. Jeffery. Current successes and remaining challenges in protein function prediction.Frontiers in Bioin- formatics, 3, July 2023. ISSN 2673-7647. doi: 10.3389/fbinf.2023.1222182. URL https://www.frontiersin. org/journals/bioinformatics/articles/10.3389/fbinf.2023.12...

    2023

  8. [16]

    Fowler and Stanley Fields

    Douglas M. Fowler and Stanley Fields. Deep mutational scanning: a new style of protein science. Nature Methods, 11(8):801–807, August 2014. ISSN 1548-7105. doi: 10.1038/nmeth.3027. URL https://www.nature.com/ articles/nmeth.3027. Publisher: Nature Publishing Group

    2014 doi

  9. [17]

    Antibody complementarity-determining region design using AlphaFold2 and DDG predictor

    Takafumi Ueki and Masahito Ohue. Antibody complementarity-determining region design using AlphaFold2 and DDG predictor. The Journal of Supercomputing, 80(9):11989–12002, June 2024. ISSN 1573-0484. doi: 10.1007/s11227-023-05887-9. URL https://doi.org/10.1007/s11227-023-05887-9

    2024 doi

  10. [18]

    ToxDL 2.0: Protein toxicity prediction using a pretrained language model and graph neural networks

    Lin Zhu, Yi Fang, Shuting Liu, Hong-Bin Shen, Wesley De Neve, and Xiaoyong Pan. ToxDL 2.0: Protein toxicity prediction using a pretrained language model and graph neural networks. Computational and Structural Biotechnology Journal, 27:1538–1549, April 2025. ISSN 2001-0370. doi...

    2025 doi

  11. [19]

    Towards Accurate and Efficient Binding Affinity Prediction

    Saro Passaro, Gabriele Corso, Jeremy Wohlwend, Mateo Reveiz, Stephan Thaler, Vignesh Ram Somnath, Noah Getz, Tally Portnoi, Julien Roy, Hannes Stark, David Kwabi-Addo, Dominique Beaini, Tommi Jaakkola, and Regina Barzilay. Towards Accurate and Efficient Binding Affinity Prediction

  12. [20]

    Sriram Kosuri and George M. Church. Large-scale de novo DNA synthesis: technologies and applications. Nature Methods, 11(5):499–507, May 2014. ISSN 1548-7105. doi: 10.1038/nmeth.2918. URL https://www.nature. com/articles/nmeth.2918. Publisher: Nature Publishing Group

    2014 doi

  13. [21]

    Anderson, Karen W

    Po-E Li, Chien-Chi Lo, Joseph J. Anderson, Karen W. Davenport, Kimberly A. Bishop-Lilly, Yan Xu, Sanaa Ahmed, Shihai Feng, Vishwesh P. Mokashi, and Patrick S.G. Chain. Enabling the democratization of the genomics revolution with a fully integrated web-based bioinformatics plat...

  14. [22]

    PHIStruct: improving phage–host interaction prediction at low sequence similarity settings using structure-aware protein embeddings.Bioinformatics, 41(1):btaf016, January 2025

    Mark Edward M Gonzales, Jennifer C Ureta, and Anish M S Shrestha. PHIStruct: improving phage–host interaction prediction at low sequence similarity settings using structure-aware protein embeddings.Bioinformatics, 41(1):btaf016, January 2025. ISSN 1367-4811. doi: 10.1093/bioin...

    2025 doi

  15. [23]

    SpatialPPIv2: Enhancing protein–protein interaction prediction through graph neural networks with protein language models

    Wenxing Hu and Masahito Ohue. SpatialPPIv2: Enhancing protein–protein interaction prediction through graph neural networks with protein language models. Computational and Structural Biotechnology Journal, 27:508–518, January 2025. ISSN 2001-0370. doi: 10.1016/j.csbj.2025.01.02...

    2025 doi

  16. [24]

    Kuchibhatla, Westley A

    Durga B. Kuchibhatla, Westley A. Sherman, Betty Y . W. Chung, Shelley Cook, Georg Schneider, Birgit Eisenhaber, and David G. Karlin. Powerful sequence similarity search methods and in-depth manual analyses can identify remote homologs in many apparently “orphan” viral proteins...

    2014 doi

  17. [25]

    Mmseqs2: sensitive protein sequence searching for the analysis of massive data sets

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

    2017 doi

  18. [26]

    Altschul, Warren Gish, Webb Miller, Eugene W

    Stephen F. Altschul, Warren Gish, Webb Miller, Eugene W. Myers, and David J. Lipman. Basic local align- ment search tool. Journal of Molecular Biology , 215(3):403–410, 1990. ISSN 0022-2836. doi: https: //doi.org/10.1016/S0022-2836(05)80360-2. URL https://www.sciencedirect.com...

    1990 doi

  19. [27]

    Rodney Brister, Danso Ako-adjei, Yiming Bao, and Olga Blinkova

    J. Rodney Brister, Danso Ako-adjei, Yiming Bao, and Olga Blinkova. Ncbi viral genomes resource. Nucleic Acids Research, 43(D1):D571–D577, 11 2014. ISSN 0305-1048. doi: 10.1093/nar/gku1207. URL https: //doi.org/10.1093/nar/gku1207

    2014 doi

  20. [28]

    The universal protein resource (uniprot) in 2010

    The UniProt Consortium. The universal protein resource (uniprot) in 2010. Nucleic Acids Research, 38(suppl_1): D142–D148, 10 2009. ISSN 0305-1048. doi: 10.1093/nar/gkp846. URL https://doi.org/10.1093/nar/ gkp846

    2010 doi

  21. [29]

    ColabFold: making protein folding accessible to all

    Milot Mirdita, Konstantin Schütze, Yoshitaka Moriwaki, Lim Heo, Sergey Ovchinnikov, and Martin Steinegger. ColabFold: making protein folding accessible to all. Nature Methods, 19(6):679–682, June 2022. ISSN 1548-7105. doi: 10.1038/s41592-022-01488-1. URL https://www.nature.com...

    2022 doi

  22. [30]

    Predicting protein–protein interactions through sequence-based deep learning

    Somaye Hashemifar, Behnam Neyshabur, Aly A Khan, and Jinbo Xu. Predicting protein–protein interactions through sequence-based deep learning. Bioinformatics, 34(17):i802–i810, September 2018. ISSN 1367-4803. doi: 10.1093/bioinformatics/bty573. URL https://doi.org/10.1093/bioinf...

    2018 doi

  23. [31]

    Michael Gromiha

    Rahul Nikam, Kumar Yugandhar, and M. Michael Gromiha. Deep learning-based method for predicting and classifying the binding affinity of protein-protein complexes. Biochimica et Biophysica Acta (BBA) - Proteins and Proteomics, 1871(6):140948, November 2023. ISSN 1570-9639. doi:...

    2023

  24. [32]

    Improved deep learning prediction of antigen-antibody interactions

    Mu Gao and Jeffrey Skolnick. Improved deep learning prediction of antigen-antibody interactions. Proceedings of the National Academy of Sciences of the United States of America, 121(41):e2410529121, October 2024. ISSN 1091-6490. doi: 10.1073/pnas.2410529121

    2024 doi

  25. [33]

    Thadani, Sarah Gurev, Pascal Notin, Noor Youssef, Nathan J

    Nicole N. Thadani, Sarah Gurev, Pascal Notin, Noor Youssef, Nathan J. Rollins, Daniel Ritter, Chris Sander, Yarin Gal, and Debora S. Marks. Learning from prepandemic data to forecast viral escape. Nature, 622(7984): 818–825, October 2023. ISSN 1476-4687. doi: 10.1038/s41586-02...

    2023 doi

  26. [34]

    Maginnis

    Melissa S. Maginnis. Virus–receptor interactions: The key to cellular invasion. Journal of Molecular Biology, 430(17):2590–2611, 2018. ISSN 0022-2836. doi: https://doi.org/10.1016/j.jmb.2018.06.024. URL https: //www.sciencedirect.com/science/article/pii/S0022283618306302

    2018 doi

  27. [35]

    Pitfalls of machine learning models for protein–protein interaction networks

    Loïc Lannelongue and Michael Inouye. Pitfalls of machine learning models for protein–protein interaction networks. Bioinformatics, 40(2):btae012, 01 2024. ISSN 1367-4811. doi: 10.1093/bioinformatics/btae012. URL https://doi.org/10.1093/bioinformatics/btae012

    2024 doi

  28. [36]

    Rcsb protein data bank: biological macromolecular structures enabling research and education in fundamental biology, biomedicine, biotechnology and energy

    Stephen K Burley, Helen M Berman, Charmi Bhikadiya, Chunxiao Bi, Li Chen, Luigi Di Costanzo, Cole Christie, Ken Dalenberg, Jose M Duarte, Shuchismita Dutta, Zukang Feng, Sutapa Ghosh, David S Goodsell, Rachel K Green, Vladimir Guranovi´c, Dmytro Guzenko, Brian P Hudson, Tara K...

    2018

  29. [37]

    Ballard, Joshua Bambrick, Sebastian W

    Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J. Ballard, Joshua Bambrick, Sebastian W. Bodenstein, David A. Evans, Chia-Chun Hung, Michael O’Neill, David Reiman, Kathryn Tunyasuvunakool, Zachar...

    2024

  30. [38]

    AF3Complex Yields Improved Structural Predictions of Protein Com- plexes, March 2025

    Jonathan Feldman and Jeffrey Skolnick. AF3Complex Yields Improved Structural Predictions of Protein Com- plexes, March 2025. URL https://www.biorxiv.org/content/10.1101/2025.02.27.640585v1. Pages: 2025.02.27.640585 Section: New Results

    2025 doi

  31. [39]

    Benchmarking AlphaFold3’s protein-protein complex accuracy and machine learning prediction reliability for binding free energy changes upon mutation

    JunJie Wee and Guo-Wei Wei. Benchmarking AlphaFold3’s protein-protein complex accuracy and machine learning prediction reliability for binding free energy changes upon mutation. ArXiv, page arXiv:2406.03979v1, June 2024. ISSN 2331-8422. URL https://www.ncbi.nlm.nih.gov/pmc/art...

    2024 arXiv

  32. [40]

    H. Zhao, C. Velez, A. Navarene, A. Saha, J. Feldman, J. Skolnick, D. Murray, and B. Honig. Combining structural modeling and deep learning to calculate the e. coli protein interactome and functional networks. bioRxiv, 2025. doi: 10.1101/2025.05.07.652715. URL https://www.biorx...

    2025 doi

  33. [41]

    Parks, and Jeffrey Skolnick

    Mu Gao, Davi Nakajima An, Jerry M. Parks, and Jeffrey Skolnick. AF2Complex predicts direct physical interac- tions in multimeric proteins with deep learning.Nature Communications, 13(1):1744, April 2022. ISSN 2041-1723. doi: 10.1038/s41467-022-29394-2. URL https://www.nature.c...

    2022 doi

  34. [42]

    Baptista-Hon, Xiaohong Helena Yang, Kanmin Xue, Wa Hou Tai, Zeyu Jiang, Linling Cheng, Manson Fok, Johnson Yiu-Nam Lau, Shengyong Yang, Ligong Lu, Ping Zhang, and Kang Zhang

    Guangyu Wang, Xiaohong Liu, Kai Wang, Yuanxu Gao, Gen Li, Daniel T. Baptista-Hon, Xiaohong Helena Yang, Kanmin Xue, Wa Hou Tai, Zeyu Jiang, Linling Cheng, Manson Fok, Johnson Yiu-Nam Lau, Shengyong Yang, Ligong Lu, Ping Zhang, and Kang Zhang. Deep-learning-enabled protein–prot...

    2007

  35. [43]

    Pinder: The protein interaction dataset and evaluation resource

    Daniel Kovtun, Mehmet Akdel, Alexander Goncearenco, Guoqing Zhou, Graham Holt, David Baugher, Dejun Lin, Yusuf Adeshina, Thomas Castiglione, Xiaoyun Wang, Céline Marquet, Matt McPartlon, Tomas Geffner, Gabriele Corso, Hannes Stärk, Zachary Carpenter, Emine Kucukbenli, Michael ...

    2024 doi

  36. [44]

    Phillips, Jeffrey Chang, Serafina Nieves, Anne A

    Alief Moulana, Thomas Dupic, Angela M. Phillips, Jeffrey Chang, Serafina Nieves, Anne A. Roffler, Allison J. Greaney, Tyler N. Starr, Jesse D. Bloom, and Michael M. Desai. Compensatory epistasis maintains ACE2 affinity in SARS-CoV-2 Omicron BA.1. Nature Communications, 13(1):7...

    2022

  37. [45]

    Recent progress and future challenges in structure-based protein-protein interaction prediction

    Rongqing Yuan, Jing Zhang, Jian Zhou, and Qian Cong. Recent progress and future challenges in structure-based protein-protein interaction prediction. Molecular Therapy, 33(5):2252–2268, 2025. ISSN 1525-0016. doi: https://doi.org/10.1016/j.ymthe.2025.04.003. URL https://www.sci...

    2025 doi

  38. [46]

    Doherty, Nathan Price, Daniel Bellieny-Rabelo, Yong K

    Jason Nomburg, Erin E. Doherty, Nathan Price, Daniel Bellieny-Rabelo, Yong K. Zhu, and Jennifer A. Doudna. Birth of protein folds and functions in the virome. Nature, 633(8030):710–717, September 2024. ISSN 1476-4687. doi: 10.1038/s41586-024-07809-y. URL https://www.nature.com...

    2024 doi

  39. [47]

    Current progress, challenges, and future perspectives of language models for protein representation and protein design

    Tao Huang and Yixue Li. Current progress, challenges, and future perspectives of language models for protein representation and protein design. The Innovation , 4(4):100446, 2023. ISSN 2666-6758. doi: https://doi.org/10.1016/j.xinn.2023.100446. URL https://www.sciencedirect.co...

    2023

  40. [48]

    Five protein-design questions that still challenge AI

    Sara Reardon. Five protein-design questions that still challenge AI. Nature, 635(8037):246–248, November 2024. doi: 10.1038/d41586-024-03595-9. URL https://www.nature.com/articles/d41586-024-03595-9 . Bandiera_abtest: a Cg_type: Technology Feature Publisher: Nature Publishing ...

    2024 doi

  41. [49]

    Artificial intelligence challenges in the face of biological threats: emerging catastrophic risks for public health

    Renan Chaves de Lima, Lucas Sinclair, Ricardo Megger, Magno Alessandro Guedes Maciel, Pedro Fernando da Costa Vasconcelos, and Juarez Antônio Simões Quaresma. Artificial intelligence challenges in the face of biological threats: emerging catastrophic risks for public health. F...

  42. [50]

    Leyma P. De Haro. Biosecurity Risk Assessment for the Use of Artificial Intelligence in Synthetic Biology.Applied Biosafety: Journal of the American Biological Safety Association, 29(2):96–107, June 2024. ISSN 1535-6760. doi: 10.1089/apb.2023.0031. URL https://www.ncbi.nlm.nih...

    2024

  43. [51]

    Hummer, Constantin Schneider, Lewis Chinery, and Charlotte M

    Alissa M. Hummer, Constantin Schneider, Lewis Chinery, and Charlotte M. Deane. Investigating the volume and diversity of data needed for generalizable antibody–antigen G prediction. Nature Computational Science, pages 1–13, July 2025. ISSN 2662-8457. doi: 10.1038/s43588-025-00...

    2025 doi

  44. [52]

    Cooperation in the time of covid

    Jade Butterworth, David Smerdon, Roy Baumeister, and William von Hippel. Cooperation in the time of covid. Perspectives on Psychological Science, 19(4):640–651, 2024. doi: 10.1177/17456916231178719. URL https://doi.org/10.1177/17456916231178719. PMID: 37384624

    2024 doi

  45. [53]

    McCormick

    Cynthia G. McCormick. Regulatory challenges for new formulations of controlled substances in today’s environment. Drug and Alcohol Dependence , 83:S63–S67, 2006. ISSN 0376-8716. doi: https://doi. org/10.1016/j.drugalcdep.2006.02.001. URL https://www.sciencedirect.com/science/a...

    2006 doi

  46. [54]

    Sacks, Hala H

    Leonard V . Sacks, Hala H. Shamsuddin, Yuliya I. Yasinskaya, Khaled Bouri, Michael L. Lanthier, and Rachel E. Sherman. Scientific and regulatory reasons for delay and denial of fda approval of initial applications for new drugs, 2000-2012. JAMA, 311(4):378–384, 01 2014. ISSN 0...

  47. [2016]

    doi: 10.1093/nar/gkw1027

    ISSN 0305-1048. doi: 10.1093/nar/gkw1027. URL https://doi.org/10.1093/nar/gkw1027. 8 Resilient Biosecurity

    doi

  48. [2023]

    doi: 10.1038/s41591-023-02483-5

    ISSN 1546-170X. doi: 10.1038/s41591-023-02483-5. URL https://www.nature.com/articles/ s41591-023-02483-5 . Publisher: Nature Publishing Group

    doi

  49. [2024]

    doi: 10.3389/frai.2024.1382356

    ISSN 2624-8212. doi: 10.3389/frai.2024.1382356. URL https://www.frontiersin.org/journals/ artificial-intelligence/articles/10.3389/frai.2024.1382356/full. Publisher: Frontiers

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.