pith. sign in

arxiv: 2606.17127 · v1 · pith:BKKGSZRSnew · submitted 2026-06-15 · 🧬 q-bio.QM · cs.AI· cs.LG

Agentic Discovery of Non-Canonical Antimicrobial Peptides with AMPGAN v3

Jay Jung , Xiaohan Zhang , Shenghan Song , Mahmoud Sayedahmed , Chijian Xiang , Yunong Xu , Ahmed AbdelKhalek , Severin T. Schneebeli
show 3 more authors
Matthew J. Wargo Jianing Li Safwan Wshah
This is my paper

Pith reviewed 2026-06-27 02:32 UTC · model grok-4.3

classification 🧬 q-bio.QM cs.AIcs.LG
keywords antimicrobial peptidesgenerative adversarial networksD-amino acidspeptide modificationsmulti-agent systemsantibiotic resistancein vitro validationAMP discovery
0
0 comments X

The pith

AMPGAN v3 uses two separate discriminators to generate non-canonical antimicrobial peptides that include D-amino acids and show activity in lab tests.

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

The paper presents AMPGAN v3 as a conditional GAN that generates antimicrobial peptides containing D-amino acids and N/C-terminal modifications. It improves on earlier models by splitting supervision between one discriminator that enforces realistic sequences and another that pushes toward predicted activity, which stabilizes training. Five generated candidates from different structural classes were synthesized and tested; two inhibited Gram-positive bacteria, and the strongest reached an MIC of 8 μg/mL against B. subtilis. The authors also introduce PepCraft, a multi-agent system that coordinates generation, filtering, and verification steps, with its rankings matching the lab results. The work therefore demonstrates one concrete way generative and agentic AI can be combined for therapeutic peptide design against antibiotic resistance.

Core claim

AMPGAN v3 is a multi-objective conditional GAN whose vocabulary includes D-amino acids and N/C-terminus modifications such as amidation. By assigning adversarial supervision to one discriminator and activity-aware supervision to a second, the model achieves greater training stability and higher scores on external activity classifiers than prior generative AMP models. Five candidates spanning three structural classes were advanced to in vitro testing; two proved active against Gram-positive strains, with the best candidate recording an MIC of 8 μg/mL against B. subtilis. PepCraft, a multi-agent orchestration framework, produces prioritization decisions that align with these experimental outco

What carries the argument

The dual-discriminator architecture of AMPGAN v3 that isolates adversarial sequence realism from activity-aware guidance.

If this is right

  • The model can now propose peptides that contain non-natural amino acids and terminal modifications, moving generative design closer to molecules that can be developed as drugs.
  • Training remains stable even when the output space includes chemical modifications, removing a previous barrier for conditional GANs in peptide design.
  • A subset of generated sequences shows measurable activity in vitro, providing direct evidence that the dual-discriminator approach can produce functional molecules.
  • PepCraft demonstrates that an agentic layer can coordinate generation, filtering, and experimental prioritization in a single workflow whose outputs match laboratory results.
  • The combined generative-plus-agentic pipeline supplies an end-to-end example of AI-assisted therapeutic peptide discovery on a modest but real scale.

Where Pith is reading between the lines

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

  • The same separation of realism and activity signals could be applied to generative models for other peptide or protein properties beyond antimicrobial activity.
  • If the external classifiers continue to correlate with lab results at larger scale, the method could accelerate screening of chemical space that is otherwise expensive to explore experimentally.
  • Extending PepCraft-style orchestration to additional specialized agents might further reduce the manual steps required between in silico generation and in vitro confirmation.
  • Success with Gram-positive activity suggests the framework could be retrained or conditioned on different bacterial targets or resistance profiles to address additional clinical needs.

Load-bearing premise

External activity classifiers reliably identify which generated sequences will exhibit real antimicrobial activity when synthesized and tested in the laboratory.

What would settle it

Synthesizing and testing a larger panel of AMPGAN v3 outputs in standardized MIC assays and finding that few or none inhibit the target Gram-positive strains at the concentrations predicted by the classifiers.

Figures

Figures reproduced from arXiv: 2606.17127 by Ahmed AbdelKhalek, Chijian Xiang, Jay Jung, Jianing Li, Mahmoud Sayedahmed, Matthew J. Wargo, Safwan Wshah, Severin T. Schneebeli, Shenghan Song, Xiaohan Zhang, Yunong Xu.

Figure 1
Figure 1. Figure 1: AMPGAN-v3 architecture. The generator Gθ maps a latent vector z ∼ N (0, I) and condition c = (species − s, MIC − m, length − ℓ) to a sequence of token logits via FiLM-based conditioning, transposed convolutions, and a Transformer encoder. Two discriminators are trained jointly: Dadv classifies real vs. generated sequences unconditionally, whereas Dmic regresses MIC values from real sequences conditioned on… view at source ↗
Figure 2
Figure 2. Figure 2: Agentic workflow for AMP discovery. A Planning Agent translates user objectives into natural-language instructions and send them to specialized executors with tools: Generating Agent, Filtering Agent and Verifying Agent. Each executor returns a report, and Planning Agent iterates until the objective is met. The selected executor e executes its designated tools along with the instruction prompt generated by… view at source ↗
Figure 3
Figure 3. Figure 3: Sequence property distributions for generated and training AMPs. Per-residue frequencies (Top), net charge at pH 7 (center), and GRAVY hydrophobicity (Bottom), Generated sequences concentrate within AMP-active ranges charge in [+1, +9]; GRAVY in [-1, +1]).Top: per-residue relative frequencies closely match the training distribution, except for Arg (R) and Trp (W), which are overrepresented by ∼ 0.025. AMPG… view at source ↗
Figure 4
Figure 4. Figure 4: Multi-agent final report. The Planning Agent synthesizes BLAST hits from the Verifying Agent into per-candidate findings, aggregate observations, and prioritization recommendations. Representative trajectory can be found in Section C.5 K = 20 [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Total tool calls per run as a function of task size. Each setting was run across N = 10 independent seeds at K ∈ {5, 10, 20} requested candidates. Lines show per-seed mean tool calls; shaded bands show 95% bootstrap confidence intervals; individual points show per-seed counts. running the Verification pipeline on the five in vitro can￾didates from Section 5.1. The Verifying Agent retrieves BLAST (Camacho e… view at source ↗
read the original abstract

Antimicrobial resistance causes to over a million deaths annually. Antimicrobial peptides (AMPs) are a promising solution, but generative AMP models are not yet ready to design peptides with non-natural amino acids and/or chemical modifications, which are essential for real-world peptide drugs. We present AMPGAN v3, a multi-objective conditional GAN that expands the generative vocabulary to D-amino acids and N/C-terminus modifications such as amidation. By separating adversarial and activity-aware supervision across two specialized discriminators, AMPGAN v3 substantially improves training stability and outperforms prior generative AMP models on external classifiers. We validated five candidates spanning three structural classes in vitro; two showed activity against Gram-positive strains, with the best candidate reaching MIC 8 {\mu}g/mL against B. subtilis. To support downstream curation, we further present PepCraft, a multi-agent framework for end-to-end AMP discovery in which a Planning Agent orchestrates specialized executors for generation, filtering, and verification. Its prioritization recommendations align with our in vitro outcomes. Together, these contributions let us examine, on a small but real scale, how generative and agentic AI compose in therapeutic peptide discovery. Code: https://github.com/marszzibros/AMPGANv3

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 1 minor

Summary. The manuscript introduces AMPGAN v3, a multi-objective conditional GAN that generates antimicrobial peptides incorporating D-amino acids and N/C-terminal modifications. It claims that separating adversarial and activity-aware supervision into two specialized discriminators improves training stability and yields better performance than prior generative AMP models when evaluated on external classifiers. Five generated candidates across three structural classes were tested in vitro, with two showing activity against Gram-positive bacteria (best MIC of 8 μg/mL vs. B. subtilis). The work also presents PepCraft, a multi-agent framework for end-to-end AMP discovery whose prioritization aligns with the in vitro results, and releases code at the provided GitHub link.

Significance. If the performance and validation claims hold after detailed reporting, the work would contribute to extending generative peptide design beyond canonical amino acids, which is relevant for therapeutic development, and would illustrate a concrete composition of generative and agentic methods. The open release of code is a clear strength supporting reproducibility.

major comments (3)
  1. [Abstract] Abstract: the claim that separating supervision across two discriminators 'substantially improves training stability and outperforms prior generative AMP models on external classifiers' is presented without any quantitative metrics, ablation results, dataset sizes, or statistical comparisons, which is load-bearing for the central methodological contribution.
  2. [Abstract] Abstract: the in vitro validation reports activity for only two of five candidates but provides no information on experimental controls, replicates, error bars, statistical tests, or head-to-head comparison against peptides generated by prior models, limiting support for the claim that the generative approach produces functional non-canonical AMPs.
  3. [Abstract] Abstract: outperformance and candidate selection both rely on external classifiers as proxies for antimicrobial activity, yet no correlation data, ablation on classifier choice, or independent validation of those classifiers against the described lab setting is supplied.
minor comments (1)
  1. [Abstract] Abstract: 'causes to over a million deaths annually' contains a grammatical error and should read 'causes over a million deaths annually'.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the abstract to better support the claims with quantitative details and experimental context where possible.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that separating supervision across two discriminators 'substantially improves training stability and outperforms prior generative AMP models on external classifiers' is presented without any quantitative metrics, ablation results, dataset sizes, or statistical comparisons, which is load-bearing for the central methodological contribution.

    Authors: The quantitative metrics supporting improved training stability (e.g., loss variance reductions), performance gains on external classifiers, ablation results, dataset sizes, and statistical comparisons are reported in the main text (Results and Methods sections). To make the abstract self-contained, we will revise it to incorporate key quantitative values and references to these analyses. revision: yes

  2. Referee: [Abstract] Abstract: the in vitro validation reports activity for only two of five candidates but provides no information on experimental controls, replicates, error bars, statistical tests, or head-to-head comparison against peptides generated by prior models, limiting support for the claim that the generative approach produces functional non-canonical AMPs.

    Authors: The full manuscript describes the in vitro methods, including positive/negative controls, triplicate replicates with error bars, and statistical tests applied to MIC values. We will update the abstract to summarize these elements. Head-to-head comparisons against peptides from prior models were not performed, as the study prioritized validation of non-canonical candidates; this remains a limitation. revision: partial

  3. Referee: [Abstract] Abstract: outperformance and candidate selection both rely on external classifiers as proxies for antimicrobial activity, yet no correlation data, ablation on classifier choice, or independent validation of those classifiers against the described lab setting is supplied.

    Authors: The manuscript reports validation of the external classifiers on independent benchmark sets and notes alignment between classifier scores and the in vitro outcomes for PepCraft prioritization. We will revise the abstract to reference this correlation and validation. Limited ablation across classifier variants is discussed in the evaluation but can be highlighted more explicitly. revision: yes

Circularity Check

0 steps flagged

No circularity; claims rest on model architecture and external validation without self-referential reductions.

full rationale

The paper presents AMPGAN v3 as a conditional GAN using two specialized discriminators for adversarial and activity-aware supervision, with claims of improved stability and outperformance on external classifiers, followed by in vitro validation of five synthesized candidates (two active). No equations, derivations, fitted parameters renamed as predictions, or self-citation load-bearing steps appear in the abstract or described content. The in vitro MIC results and PepCraft framework provide independent external grounding rather than reducing to model inputs by construction. The derivation chain is self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review yields minimal ledger entries. No free parameters, invented entities, or additional axioms are explicitly stated beyond standard GAN training assumptions.

axioms (1)
  • domain assumption Separating adversarial and activity-aware supervision across two discriminators improves training stability
    Invoked as the mechanism for the reported stability gain.

pith-pipeline@v0.9.1-grok · 5799 in / 1155 out tokens · 50532 ms · 2026-06-27T02:32:33.294904+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

45 extracted references · 23 canonical work pages

  1. [1]

    Nucleic Acids Research , author=

    DBAASP v3: database of antimicrobial/cytotoxic activity and structure of peptides as a resource for development of new therapeutics , volume=. Nucleic Acids Research , author=. 2020 , month=. doi:https://doi.org/10.1093/nar/gkaa991 , number=

    work page doi:10.1093/nar/gkaa991 2020

  2. [2]

    Journal of Chemical Information and Modeling , author=

    AMPGAN v2: Machine Learning-Guided Design of Antimicrobial Peptides , volume=. Journal of Chemical Information and Modeling , author=. 2021 , month=. doi:https://doi.org/10.1021/acs.jcim.0c01441 , number=

    work page doi:10.1021/acs.jcim.0c01441 2021

  3. [3]

    A Generative Approach toward Precision Antimicrobial Peptide Design , DOI=

    Ferrell, Jonathon B and Remington, Jacob M and Van, Colin M and Sharafi, Mona and Reem Aboushousha and Janssen-Heininger, Yvonne and Schneebeli, Severin T and Wargo, Matthew J and Safwan Wshah and Li, Jianing , year=. A Generative Approach toward Precision Antimicrobial Peptide Design , DOI=. bioRxiv (Cold Spring Harbor Laboratory) , publisher=

  4. [4]

    Current economic and regulatory challenges in developing antibiotics for Gram-negative bacteria , volume=

    Gargate, Nupur and Laws, Mark and Rahman, Khondaker Miraz , year=. Current economic and regulatory challenges in developing antibiotics for Gram-negative bacteria , volume=. npj Antimicrobials and Resistance , publisher=. doi:https://doi.org/10.1038/s44259-025-00123-1 , number=

    work page doi:10.1038/s44259-025-00123-1

  5. [5]

    Challenges of antibacterial discovery , volume=

    Silver, Lynn L , year=. Challenges of antibacterial discovery , volume=. Clinical microbiology reviews , publisher=. doi:https://doi.org/10.1128/CMR.00030-10 , number=

    work page doi:10.1128/cmr.00030-10

  6. [6]

    Antimicrobial resistance , url=

    WHO , year=. Antimicrobial resistance , url=

  7. [7]

    The Lancet , author =

    Global Burden of Bacterial Antimicrobial Resistance in 2019: a Systematic Analysis , volume=. The Lancet , author =. 2022 , month=. doi:https://doi.org/10.1016/S0140-6736(21)02724-0 , number=

    work page doi:10.1016/s0140-6736(21)02724-0 2019

  8. [8]

    Antimicrobial Peptides: Opportunities and Challenges in Overcoming Resistance , volume=

    Bucataru, Cezara and Ciobanasu, Corina , year=. Antimicrobial Peptides: Opportunities and Challenges in Overcoming Resistance , volume=. doi:https://doi.org/10.1016/j.micres.2024.127822 , journal=

    work page doi:10.1016/j.micres.2024.127822 2024

  9. [9]

    Nature Biomedical Engineering , author=

    Deep-learning-enabled antibiotic discovery through molecular de-extinction , DOI=. Nature Biomedical Engineering , author=. 2024 , month=

    2024

  10. [10]

    Nature Communications , author=

    Discovering highly potent antimicrobial peptides with deep generative model HydrAMP , volume=. Nature Communications , author=. 2023 , month=. doi:https://doi.org/10.1038/s41467-023-36994-z , number=

    work page doi:10.1038/s41467-023-36994-z 2023

  11. [11]

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

    Evolutionary-scale prediction of atomic-level protein structure with a language model , volume=. Science , author=. 2023 , month=. doi:https://doi.org/10.1126/science.ade2574 , number=

    work page doi:10.1126/science.ade2574 2023

  12. [12]

    Science Advances , author=

    Discovery of antimicrobial peptides with notable antibacterial potency by an LLM-based foundation model , volume=. Science Advances , author=. 2025 , month=. doi:https://doi.org/10.1126/sciadv.ads8932 , number=

    work page doi:10.1126/sciadv.ads8932 2025

  13. [13]

    Briefings in Bioinformatics , author=

    MOFormer: navigating the antimicrobial peptide design space with Pareto-based multi-objective transformer , volume=. Briefings in Bioinformatics , author=. 2025 , month=. doi:https://doi.org/10.1093/bib/bbaf376 , number=

    work page doi:10.1093/bib/bbaf376 2025

  14. [14]

    2025 , eprint=

    AutoLabs: Cognitive Multi-Agent Systems with Self-Correction for Autonomous Chemical Experimentation , author=. 2025 , eprint=. doi:https://doi.org/10.48550/arXiv.2509.25651 , journal=

    work page doi:10.48550/arxiv.2509.25651 2025

  15. [15]

    2026 , doi =

    Xu, Mingyang and Yan, Jiaxian and Feng, Runchuan and Cai, Qingran and Zhang, Peng and Zhao, Chuang and He, Cao and Wei, Zongting and Li, Jianping and Lin, Shiyi and Dong, Hongyu and Jin, Ruofan and Hou, Tingjun and Liu, Qi and Zhang, Zaixi , title =. 2026 , doi =. https://www.biorxiv.org/content/early/2026/04/14/2026.04.11.716807.full.pdf , journal =

    2026

  16. [16]

    Mode Collapse in Generative Adversarial Networks: An Overview , year=

    Kossale, Youssef and Airaj, Mohammed and Darouichi, Aziz , booktitle=. Mode Collapse in Generative Adversarial Networks: An Overview , year=

  17. [17]

    Verkuil, Robert and Kabeli, Ori and Du, Yilun and Wicky, Basile I. M. and Milles, Lukas F. and Dauparas, Justas and Baker, David and Ovchinnikov, Sergey and Sercu, Tom and Rives, Alexander , title =. 2022 , doi =. https://www.biorxiv.org/content/early/2022/12/22/2022.12.21.521521.full.pdf , journal =

    2022

  18. [18]

    2026 , doi =

    Fallahpour, Adibvafa and Seyed-Ahmadi, Arman and Idehpour, Parsa and Ibrahim, Omar and Gupta and others , title =. 2026 , doi =. https://www.biorxiv.org/content/early/2026/03/20/2026.03.19.712954.full.pdf , journal =

    2026

  19. [19]

    Nucleic Acids Research , volume =

    Wang, Guangshun and Schmidt, Cindy and Li, Xia and Wang, Zhe , title =. Nucleic Acids Research , volume =. 2026 , month =. doi:10.1093/nar/gkaf860 , url =

    work page doi:10.1093/nar/gkaf860 2026

  20. [20]

    A Review of Antimicrobial Peptides: Structure, Mechanism of Action, and Molecular Optimization Strategies , volume=

    Ma, Xu and Wang, Qiang and Ren, Kexin and Xu, Tongtong and Zhang, Zigang and Xu, Meijuan and Rao, Zhiming and Zhang, Xian , year=. A Review of Antimicrobial Peptides: Structure, Mechanism of Action, and Molecular Optimization Strategies , volume=. Fermentation , publisher=. doi:https://doi.org/10.3390/fermentation10110540 , number=

    work page doi:10.3390/fermentation10110540

  21. [21]

    Antimicrobial peptides: from discovery to developmental applications , volume=

    Zhang, Qi , editor=. Antimicrobial peptides: from discovery to developmental applications , volume=. Applied and Environmental Microbiology , publisher=. 2025 , month=. doi:https://doi.org/10.1128/aem.02115-24 , number=

    work page doi:10.1128/aem.02115-24 2025

  22. [22]

    Kinetics of antimicrobial peptide activity measured on individual bacterial cells using high-speed atomic force microscopy , volume=

    Fantner, Georg E and Barbero, Roberto J and Gray, David S and Belcher, Angela M , year=. Kinetics of antimicrobial peptide activity measured on individual bacterial cells using high-speed atomic force microscopy , volume=. Nature nanotechnology , publisher=. doi:https://doi.org/10.1038/nnano.2010.29 , number=

    work page doi:10.1038/nnano.2010.29 2010

  23. [23]

    ACS Omega , volume =

    Tucs, Andrejs and Tran, Duy Phuoc and Yumoto, Akiko and Ito, Yoshihiro and Uzawa, Takanori and Tsuda, Koji , title =. ACS Omega , volume =. 2020 , doi =

    2020

  24. [24]

    and Zou, Quan and Zeng, Xiangxiang and Cao, Dongsheng , title =

    Wang, Li and Liu, Yiping and Fu, Xiangzheng and Ye, Xiucai and Shi, Junfeng and Yen, Gary G. and Zou, Quan and Zeng, Xiangxiang and Cao, Dongsheng , title =. Journal of Medicinal Chemistry , volume =. 2025 , doi =

    2025

  25. [25]

    2018 , isbn =

    Guo, Jiaxian and Lu, Sidi and Cai, Han and Zhang, Weinan and Yu, Yong and Wang, Jun , title =. 2018 , isbn =

    2018

  26. [26]

    AAAI , year=

    FiLM: Visual Reasoning with a General Conditioning Layer , author=. AAAI , year=

  27. [27]

    ICML 2025 Generative AI and Biology (GenBio) Workshop , year=

    ProteinCrow: A Language Model Agent That Can Design Proteins , author=. ICML 2025 Generative AI and Biology (GenBio) Workshop , year=

    2025

  28. [28]

    2022 , month =

    Chase, Harrison , title =. 2022 , month =

    2022

  29. [29]

    Bioinformatics , author=

    Deep learning improves antimicrobial peptide recognition , volume=. Bioinformatics , author=. 2018 , month=. doi:https://doi.org/10.1093/bioinformatics/bty179 , number=

    work page doi:10.1093/bioinformatics/bty179 2018

  30. [30]

    Scientific Reports , author=

    AmPEP: Sequence-based prediction of antimicrobial peptides using distribution patterns of amino acid properties and random forest , volume=. Scientific Reports , author=. 2018 , month=. doi:https://doi.org/10.1038/s41598-018-19752-w , number=

    work page doi:10.1038/s41598-018-19752-w 2018

  31. [31]

    Ballard, Joshua Bambrick, et al

    Accurate structure prediction of biomolecular interactions with AlphaFold 3 , volume=. Nature , author=. 2024 , month=. doi:https://doi.org/10.1038/s41586-024-07487-w , number=

    work page doi:10.1038/s41586-024-07487-w 2024

  32. [32]

    FEBS Letters , author=

    Role of arginine and lysine in the antimicrobial mechanism of histone-derived antimicrobial peptides , volume=. FEBS Letters , author=. 2015 , month=. doi:https://doi.org/10.1016/j.febslet.2015.11.002 , number=

    work page doi:10.1016/j.febslet.2015.11.002 2015

  33. [33]

    SimpleFold: Folding Proteins is Simpler than You Think , booktitle =

    Yuyang Wang and Jiarui Lu and Navdeep Jaitly and Josh Susskind and Miguel Angel Bautista , year =. SimpleFold: Folding Proteins is Simpler than You Think , booktitle =

  34. [34]

    arXiv preprint arXiv:2403.05530 , year=

    Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context , author=. arXiv preprint arXiv:2403.05530 , year=

    Pith/arXiv arXiv

  35. [35]

    arXiv preprint arXiv:2312.11805 , year=

    Gemini: a family of highly capable multimodal models , author=. arXiv preprint arXiv:2312.11805 , year=

    Pith/arXiv arXiv

  36. [36]

    arXiv preprint arXiv:2403.08295 , year=

    Gemma: Open models based on gemini research and technology , author=. arXiv preprint arXiv:2403.08295 , year=

    Pith/arXiv arXiv

  37. [37]

    and Arruga, Yolanda and Petrachi, Emilia and Roig, Albert and Lucchi, Roberta and Oller-Salvia, Benjamí , TITLE =

    Lucana, Maria C. and Arruga, Yolanda and Petrachi, Emilia and Roig, Albert and Lucchi, Roberta and Oller-Salvia, Benjamí , TITLE =. Pharmaceutics , VOLUME =. 2021 , NUMBER =

    2021

  38. [38]

    Acta Naturae , author=

    Design of Stable -Helical Peptides and Thermostable Proteins in Biotechnology and Biomedicine , volume=. Acta Naturae , author=. 2016 , month=. doi:https://doi.org/10.32607/20758251-2016-8-4-70-81 , number=

    work page doi:10.32607/20758251-2016-8-4-70-81 2016

  39. [39]

    UniProtKB/Swiss-Prot , volume=

    Boutet, Emmanuel and Lieberherr, Damien and Tognolli, Michael and Schneider, Michel and Bairoch, Amos , year=. UniProtKB/Swiss-Prot , volume=. doi:https://doi.org/10.1007/978-1-59745-535-0_4 , journal=

    work page doi:10.1007/978-1-59745-535-0_4

  40. [40]

    BMC Bioinformatics , author=

    BLAST+: architecture and applications , volume=. BMC Bioinformatics , author=. 2009 , pages=

    2009

  41. [41]

    Annals of Translational Medicine , author=

    Introduction to Machine learning: k-nearest Neighbors , volume=. Annals of Translational Medicine , author=. 2016 , month=. doi:https://doi.org/10.21037/atm.2016.03.37 , number=

    work page doi:10.21037/atm.2016.03.37 2016

  42. [42]

    2026 , note =

    Gemini 3.1 Pro , howpublished =. 2026 , note =

    2026

  43. [43]

    2026 , note =

    Gemini 3.1 Flash , howpublished =. 2026 , note =

    2026

  44. [44]

    Nature , author=

    Highly Accurate Protein Structure Prediction with Alphafold , volume=. Nature , author=. 2021 , month=. doi:https://doi.org/10.1038/s41586-021-03819-2 , number=

    work page doi:10.1038/s41586-021-03819-2 2021

  45. [45]

    2025 , note =

    Maestro,. 2025 , note =

    2025