CoChat
Sign Up
Cited by 324Computational Drug Discovery MethodsChemical Synthesis and AnalysisProtein Structure and Dynamics

The DOI matches its Crossref record; 79 of the 79 references it cites were found in a registry or an index.

Design of target specific peptide inhibitors using generative deep learning and molecular dynamics simulations

Open access
Chen, Sijie, Lin, Tong, Basu, Ruchira, Ritchey, Jeremy, Wang, Shen, Luo, Yichuan, Li, Xingcan, Pei, Dehua, Kara, Levent Burak, Cheng, Xiaolin·2024·Nature Communications
DOI 10.1038/s41467-024-45766-2PMC PMC10882002

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

We introduce a computational approach for the design of target-specific peptides. Our method integrates a Gated Recurrent Unit-based Variational Autoencoder with Rosetta FlexPepDock for peptide sequence generation and binding affinity assessment. Subsequently, molecular dynamics simulations are employed to narrow down the selection of peptides for experimental assays. We apply this computational strategy to design peptide inhibitors that specifically target β -catenin and NF- κ B essential modulator. Among the twelve β -catenin inhibitors, six exhibit improved binding affinity compared to the parent peptide. Notably, the best C-terminal peptide binds β -catenin with an IC 50 of 0.010 ± 0.06 μM, which is 15-fold better than the parent peptide. For NF- κ B essential modulator, two of the four tested peptides display substantially enhanced binding compared to the parent peptide. Collectively, this study underscores the successful integration of deep learning and structure-based modeling and simulation for target specific peptide design.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41467-024-45766-2

Cites

79 of 79 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Illustrations of the Wnt/β-catenin pathway and the binding mode of β-catenin and a stapled peptide. a Wnt signal on/off pathways. Cartoons were created with BioRender.com. b Crystal structure of a stapled peptide StAX-35R bound to β-catenin viewed from the N- or C-terminus of the peptide (PDBid: 4DJS32). c Helical length comparisons of the stapled peptide with the three interacting helices of β-catenin. d Correlation plots of I_bsa vs. I_sc and rmsALL_if vs. I_sc for 20 Rosetta Design generated N- or C-terminal extensions. For each peptide, 100 conformations were sampled, with each conformation represented by a scatter on the scatterplot. The sample sizes in (d) are n = 2000. Source data are provided as a Source Data file.
Fig. 1 Illustrations of the Wnt/β-catenin pathway and the binding mode of β-catenin and a stapled peptide. a Wnt signal on/off pathways. Cartoons were created with BioRender.com. b Crystal structure of a stapled peptide StAX-35R bound to β-catenin viewed from the N- or C-terminus of the peptide (PDBid: 4DJS32). c Helical length comparisons of the stapled peptide with the three interacting helices of β-catenin. d Correlation plots of I_bsa vs. I_sc and rmsALL_if vs. I_sc for 20 Rosetta Design generated N- or C-terminal extensions. For each peptide, 100 conformations were sampled, with each conformation represented by a scatter on the scatterplot. The sample sizes in (d) are n = 2000. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 2 Overview of the peptide design methodology. a Workflow illustrating the design of β-catenin specific peptide inhibitors. The black arrow line path represents the C-terminal extension, and the green arrow line path indicates an additional fine-tuning process for C-terminal extension. b VAE-MH sequence encoding and decoding. c VAE-MH Pretrain and fine-tuning models for peptide extension.
Fig. 2 Overview of the peptide design methodology. a Workflow illustrating the design of β-catenin specific peptide inhibitors. The black arrow line path represents the C-terminal extension, and the green arrow line path indicates an additional fine-tuning process for C-terminal extension. b VAE-MH sequence encoding and decoding. c VAE-MH Pretrain and fine-tuning models for peptide extension.CC BY · PubMed Central
Fig. 3 Rosetta FlexPepDock evaluation of the VAE-MH generated N-terminal peptide extensions. a Kernel density distribution plots for interface score (I_sc), interface binding solvent accessible area (I_bsa) and root-mean square deviations of interface backbones (rmsALL_if) of N-terminally extended peptides (17–20 residues. b Correlation plots of rmsALL_if vs. I_sc and I_bsa vs. I_sc for the N-terminally extended peptides, with each point representing a peptide. For each peptide, 100 conformations are sampled for each peptide–protein complex, and their FlexPepDock scores are averaged. c Violin plots of rmsALL_if, I_sc and I_bsa for N-terminal peptide extensions generated by the VAE-MH Pretrain model in comparison with Rosetta Design. FlexPepDock scores were computed as the difference score between the extended peptide and the parent peptide. Sample sizes in (a, b) are n = 300. Coefficients of determination in (b) for I_bsa vs. I_sc are r2 = 0.4, and for I_bsa Vs. rmsALL_if are R2 = 0.2. The minima, maxima, center (median) bounds of I_bsa are 130, 283, 203 for Rosetta Design, and 118, 314, 241 for VAE-MH pretrain. Whiskers represent the interquartile range Q1 = 162 and Q3 = 237 for Rosetta Design and the interquartile range Q1 =192 and Q3 =314 for VAE-MH pretrain. The minima, maxima, center bounds of _sc are −5.8, −0.068, −2.7 for Rosetta Design, and −10, −3.1, −5.4 for VAE-MH pretrain. Whiskers represent the interquartile range Q1 =−4.5 and Q3 =−1.9 for Rosetta Design, and the interquartile range Q1 =−6.6 and Q3 =−4.1 for VAE-MH pretrain. The minima, maxima, center bounds of rmsALL_if are −0.20, 0.23, −0.071 for Rosetta Design, and −0.2, 0.24, 0.027 for VAE-MH pretrain. Whiskers represent the interquartile range Q1 = −0.13 and Q3 = −0.17 for Rosetta Design, and the interquartile range Q1 = −0.14 and Q3 = 0.15 for VAE-MH pretrain. Source data are provided as a Source Data file.
Fig. 3 Rosetta FlexPepDock evaluation of the VAE-MH generated N-terminal peptide extensions. a Kernel density distribution plots for interface score (I_sc), interface binding solvent accessible area (I_bsa) and root-mean square deviations of interface backbones (rmsALL_if) of N-terminally extended peptides (17–20 residues. b Correlation plots of rmsALL_if vs. I_sc and I_bsa vs. I_sc for the N-terminally extended peptides, with each point representing a peptide. For each peptide, 100 conformations are sampled for each peptide–protein complex, and their FlexPepDock scores are averaged. c Violin plots of rmsALL_if, I_sc and I_bsa for N-terminal peptide extensions generated by the VAE-MH Pretrain model in comparison with Rosetta Design. FlexPepDock scores were computed as the difference score between the extended peptide and the parent peptide. Sample sizes in (a, b) are n = 300. Coefficients of determination in (b) for I_bsa vs. I_sc are r2 = 0.4, and for I_bsa Vs. rmsALL_if are R2 = 0.2. The minima, maxima, center (median) bounds of I_bsa are 130, 283, 203 for Rosetta Design, and 118, 314, 241 for VAE-MH pretrain. Whiskers represent the interquartile range Q1 = 162 and Q3 = 237 for Rosetta Design and the interquartile range Q1 =192 and Q3 =314 for VAE-MH pretrain. The minima, maxima, center bounds of _sc are −5.8, −0.068, −2.7 for Rosetta Design, and −10, −3.1, −5.4 for VAE-MH pretrain. Whiskers represent the interquartile range Q1 =−4.5 and Q3 =−1.9 for Rosetta Design, and the interquartile range Q1 =−6.6 and Q3 =−4.1 for VAE-MH pretrain. The minima, maxima, center bounds of rmsALL_if are −0.20, 0.23, −0.071 for Rosetta Design, and −0.2, 0.24, 0.027 for VAE-MH pretrain. Whiskers represent the interquartile range Q1 = −0.13 and Q3 = −0.17 for Rosetta Design, and the interquartile range Q1 = −0.14 and Q3 = 0.15 for VAE-MH pretrain. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 4 MD simulations of β-catenin bound with the four experimentally tested peptides. a Workflow of our N-terminal extension design process. b MM/GBSA results of the top 10 Rosetta FlexPepDock-scored peptide extensions. Error bars represent standard deviations (from left to right in kcal/mol: 8.5, 5.9, 3, 8.2, 6.7, 8.9, 8.5, 5.4, 8.1, 2). The green horizontal line denotes the threshold cutoff of −40 kcal/mol. c Ending pose of the best N-terminally extended peptide NAL-9 bound to β-catenin after 500 ns of MD simulation. d Initial and final structures along with their superimposition of the best peptide NAL-9 bound to β-catenin during the MD simulation. e Initial and final structures along with their superimposition of the worst peptide NAL-10 bound to β-catenin during the MD simulation. Extension residues and β-catenin residues crucial to binding are labeled and represented in stick. In (b), whiskers indicate standard deviations for triplet measurements. Source data are provided as a Source Data file.
Fig. 4 MD simulations of β-catenin bound with the four experimentally tested peptides. a Workflow of our N-terminal extension design process. b MM/GBSA results of the top 10 Rosetta FlexPepDock-scored peptide extensions. Error bars represent standard deviations (from left to right in kcal/mol: 8.5, 5.9, 3, 8.2, 6.7, 8.9, 8.5, 5.4, 8.1, 2). The green horizontal line denotes the threshold cutoff of −40 kcal/mol. c Ending pose of the best N-terminally extended peptide NAL-9 bound to β-catenin after 500 ns of MD simulation. d Initial and final structures along with their superimposition of the best peptide NAL-9 bound to β-catenin during the MD simulation. e Initial and final structures along with their superimposition of the worst peptide NAL-10 bound to β-catenin during the MD simulation. Extension residues and β-catenin residues crucial to binding are labeled and represented in stick. In (b), whiskers indicate standard deviations for triplet measurements. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 5 Rosetta FlexPepDock evaluation of C-terminal peptide extensions generated by the three fine-tune models in comparison with the Pretrain model. a Workflow of the three fine-tune models for C-terminal extension, and comparisons of enrichment results analyzed by rank difference distributions for C-terminal extensions generated by various models. b Distributions of the three Rosetta FlexPepDock binding metric scores evolving during the fine-tuning process for YPEDILDKHLQRV-based extension. c Distributions of the three Rosetta FlexPepDock binding metric scores evolving during the fine-tuning process for YPEDILDKHLQRVIL-based extension. Only the top 10% Rosetta FlexPepDock scores are plotted. The green line represents a cutoff for selecting peptides subjected to MM/GBSA evaluation (rmsALL_if < −0.2 Å; I_sc < −6 REU; I_bsa > 250 Å2). Sample sizes in (a) for YPEDILDKHLQRVIL-based extension and YPEDILDKHLQRV-based extension are n = 1840 and n = 2480, respectively. Sample sizes in (b, c) are n = 400. Source data are provided as a Source Data file.
Fig. 5 Rosetta FlexPepDock evaluation of C-terminal peptide extensions generated by the three fine-tune models in comparison with the Pretrain model. a Workflow of the three fine-tune models for C-terminal extension, and comparisons of enrichment results analyzed by rank difference distributions for C-terminal extensions generated by various models. b Distributions of the three Rosetta FlexPepDock binding metric scores evolving during the fine-tuning process for YPEDILDKHLQRV-based extension. c Distributions of the three Rosetta FlexPepDock binding metric scores evolving during the fine-tuning process for YPEDILDKHLQRVIL-based extension. Only the top 10% Rosetta FlexPepDock scores are plotted. The green line represents a cutoff for selecting peptides subjected to MM/GBSA evaluation (rmsALL_if < −0.2 Å; I_sc < −6 REU; I_bsa > 250 Å2). Sample sizes in (a) for YPEDILDKHLQRVIL-based extension and YPEDILDKHLQRV-based extension are n = 1840 and n = 2480, respectively. Sample sizes in (b, c) are n = 400. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 6 Computational and experimental evaluation of our designed C-terminal peptide extensions. a Workflow of our C-terminal extension design process. b MM/GBSA energy distributions for top peptides generated by the four VAE-MH models. c MM/GBSA energies of top ten peptides for each of the two C-terminal extensions. Error bars represent standard deviations. d Crystal structures of TCF-bound β-catenin (PDBid: 1G3J, 2GL7, and 3OUW). e Sequence alignment of selected peptide extensions (with MM/GBSA-binding energy below −40 kcal/mol) for YPEDILDKHLQRV-based extension (4–9 residues). f Sequence alignment of selected peptide extensions (with MM/GBSA-binding energy below −40 kcal/mol) for YPEDILDKHLQRVIL-based extension (2–7 residues). g Initial and final structures along with their superimposition of the best peptide CAL-2 bound to β-catenin during the MD simulation. Extension residues and β-catenin residues crucial for binding are labeled and represented in stick. Sample sizes in (b) for YPEDILDKHLQRVIL-based extension and YPEDILDKHLQRV-based extension are n = 45 and n = 54, respectively. In the left panel of (b), boxes bound the interquartile range of (Q1 = −45.3, median = −41.5, Q3 = −37.8) for Pretrain, (Q1 = −43.8, median = −40.8, Q3 = −39.2) for Fine-tune1, (Q1 = −44.6, median = −43.6, Q3 = −44.6) for Fine-tune2, and (Q1 = −46.5, median = −45.4, Q3 = −43.2) for Fine-tune3; whiskers represent the most extreme values of (minima = −45.3, maxima = −37.8) for Pretrain, (minima = −47.1, maxima = −37.2) for Fine-tune1, (minima = −49.1, maxima = −39.4) for Fine-tune2, and (minima = −49.2, maxima = −39.9) for Fine-tune3. In the right panel of (b), boxes bound the interquartile range of (Q1=-47.0, median = −44.2, Q3 = −42.9) for Pretrain, (Q1 = −45.4, median=-41.4, Q3 = −39.1) for Fine-tune1, (Q1 = −48.6, median=-44.3, Q3 = −42.5) for Fine-tune2, and (Q1 = −49.8, median = −45.8, Q3 = −43.7) for Fine-tune3; whiskers represent the most extreme values of (minima = −50.2, maxima = −33.2) for Pretrain, (minima = −54.4, maxima = −34.7) for Fine-tune1, (minima = −55.3, maxima = −35.5) for Fine-tune2, and (minima = −53.3, maxima = −37.9) for Fine-tune3. In (c), whiskers indicate standard deviations for triplet measurements. Source data are provided as a Source Data file.
Fig. 6 Computational and experimental evaluation of our designed C-terminal peptide extensions. a Workflow of our C-terminal extension design process. b MM/GBSA energy distributions for top peptides generated by the four VAE-MH models. c MM/GBSA energies of top ten peptides for each of the two C-terminal extensions. Error bars represent standard deviations. d Crystal structures of TCF-bound β-catenin (PDBid: 1G3J, 2GL7, and 3OUW). e Sequence alignment of selected peptide extensions (with MM/GBSA-binding energy below −40 kcal/mol) for YPEDILDKHLQRV-based extension (4–9 residues). f Sequence alignment of selected peptide extensions (with MM/GBSA-binding energy below −40 kcal/mol) for YPEDILDKHLQRVIL-based extension (2–7 residues). g Initial and final structures along with their superimposition of the best peptide CAL-2 bound to β-catenin during the MD simulation. Extension residues and β-catenin residues crucial for binding are labeled and represented in stick. Sample sizes in (b) for YPEDILDKHLQRVIL-based extension and YPEDILDKHLQRV-based extension are n = 45 and n = 54, respectively. In the left panel of (b), boxes bound the interquartile range of (Q1 = −45.3, median = −41.5, Q3 = −37.8) for Pretrain, (Q1 = −43.8, median = −40.8, Q3 = −39.2) for Fine-tune1, (Q1 = −44.6, median = −43.6, Q3 = −44.6) for Fine-tune2, and (Q1 = −46.5, median = −45.4, Q3 = −43.2) for Fine-tune3; whiskers represent the most extreme values of (minima = −45.3, maxima = −37.8) for Pretrain, (minima = −47.1, maxima = −37.2) for Fine-tune1, (minima = −49.1, maxima = −39.4) for Fine-tune2, and (minima = −49.2, maxima = −39.9) for Fine-tune3. In the right panel of (b), boxes bound the interquartile range of (Q1=-47.0, median = −44.2, Q3 = −42.9) for Pretrain, (Q1 = −45.4, median=-41.4, Q3 = −39.1) for Fine-tune1, (Q1 = −48.6, median=-44.3, Q3 = −42.5) for Fine-tune2, and (Q1 = −49.8, median = −45.8, Q3 = −43.7) for Fine-tune3; whiskers represent the most extreme values of (minima = −50.2, maxima = −33.2) for Pretrain, (minima = −54.4, maxima = −34.7) for Fine-tune1, (minima = −55.3, maxima = −35.5) for Fine-tune2, and (minima = −53.3, maxima = −37.9) for Fine-tune3. In (c), whiskers indicate standard deviations for triplet measurements. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 7 Design of NEMO/IKK-binding peptide extensions. a Crystal structure of the NEMO/IKK complex (PDBid: 3RBT53). b Workflow illustrating the peptide design process. c Distribution of interface energy (REU) throughout fine-tuning cycles. d Visualization of peptide-NEMO binding poses for the four tested peptides. NEMO helices are depicted in gray, and peptide inhibitors are represented in distinct colors. Source data are provided as a Source Data file.
Fig. 7 Design of NEMO/IKK-binding peptide extensions. a Crystal structure of the NEMO/IKK complex (PDBid: 3RBT53). b Workflow illustrating the peptide design process. c Distribution of interface energy (REU) throughout fine-tuning cycles. d Visualization of peptide-NEMO binding poses for the four tested peptides. NEMO helices are depicted in gray, and peptide inhibitors are represented in distinct colors. Source data are provided as a Source Data file.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 7

  1. Sutskever, I., Vinyals, O. & Le, Q. V. Sequence to sequence learning with neural networks. In Proceedings of the 27th International Conference on Neural Information Processing Systems, 3104–3112 (Curran Associates, Inc., 2014).
  2. Radford, A., Narasimhan, K., Salimans, T. & Sutskever, I. Improving language understanding by generative pre-training. https://api.semanticscholar.org/CorpusID:49313245 (2018).
  3. London, N., Raveh, B., Cohen, E., Fathi, G. & Schueler-Furman, O. Rosetta flexpepdock web server—high resolution modeling of peptide–protein interactions. Nucleic Acids Res. 39, W249–W253 (2011). · DOI
  4. Doersch, C. Tutorial on variational autoencoders. Preprint at https://arxiv.org/abs/1606.05908 (2016). · arXiv
  5. Pascanu, R., Mikolov, T. & Bengio, Y. On the difficulty of training recurrent neural networks. In Proceedings of the 30th International Conference on Machine Learning (eds. Dasgupta, S., & McAllester D.) PMLR 28, 1310–1318 (JMLR.org, 2012).
  6. Chung, J., Gulcehre, C., Cho, K. & Bengio, Y. Empirical evaluation of gated recurrent neural networks on sequence modeling. Preprint at https://arxiv.org/abs/1412.3555 (2014). · arXiv
  7. Stranges, P. B. & Kuhlman, B. A comparison of successful and failed protein interface designs highlights the challenges of designing buried hydrogen bonds. Protein Sci. 22, 74–82 (2012). · DOI

Cited by

Sorted by citations

See all 324

Related

See all

Metadata

Authors
Chen, S., Lin, T., Basu, R., Ritchey, J., Wang, S., Luo, Y., Li, X., Pei, D., Kara, L. B., & Cheng, X.
Affiliations
  • The Ohio State University (US)
  • Carnegie Mellon University (US)
  • Nantong University (CN)
Corresponding author
  • Pei, Dehua
  • Kara, Levent Burak
  • Cheng, Xiaolin
ORCID
Venue
Nature Communications
Publisher
Springer Science and Business Media LLC
Published
2024-02-21
Volume
15
Issue
1
Pages
1611-1611
ISSN
2041-1723 (ISSN-L)
DOI
10.1038/s41467-024-45766-2
PMC
PMC10882002
OpenAlex
W4392091326
Cited by
324
References
78
Citations by year
2026: 121 · 2025: 158 · 2024: 44
FWCI
68.33
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • Research Square (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
  • Ohio State University (P30CA016058)
  • National Institutes of Health (1s10rr025660-01a1)
  • National Cancer Institute (CA234124, R01CA234124, P30CA016058)
  • National Institute of General Medical Sciences (R35GM122459)
  • National Center for Research Resources (S10 RR025660)
Topics
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
  • Physical Sciences › Chemistry › Organic Chemistry › Chemical Synthesis and Analysis
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
Keywords
β-catenin, NF-κB essential modulator, peptide sequence generation, binding affinity prediction, molecular dynamics simulation, structure-based drug design
MeSH terms
Deep Learning, Peptides, Protein Binding, NF-kappa B, beta Catenin, Molecular Dynamics Simulation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review