CoChat
Sign Up
Cited by 132Supramolecular Self-Assembly in MaterialsChemical Synthesis and AnalysisMachine Learning in Materials Science

The DOI matches its Crossref record; 62 of the 62 references it cites were found in a registry or an index.

Accelerating the prediction and discovery of peptide hydrogels with human-in-the-loop

Open access
Xu, Tengyan, Wang, Jiaqi, Zhao, Shuang, Chen, Dinghao, Zhang, Hongyue, Fang, Yu, Kong, Nan, Zhou, Ziao, Li, Wenbin, Wang, Huaimin·2023·Nature Communications
DOI 10.1038/s41467-023-39648-2PMC PMC10313671
Open this paper in CoChat — free accountOpen PDF

Abstract

The amino acid sequences of peptides determine their self-assembling properties. Accurate prediction of peptidic hydrogel formation, however, remains a challenging task. This work describes an interactive approach involving the mutual information exchange between experiment and machine learning for robust prediction and design of (tetra)peptide hydrogels. We chemically synthesize more than 160 natural tetrapeptides and evaluate their hydrogel-forming ability, and then employ machine learning-experiment iterative loops to improve the accuracy of the gelation prediction. We construct a score function coupling the aggregation propensity, hydrophobicity, and gelation corrector C g , and generate an 8,000-sequence library, within which the success rate of predicting hydrogel formation reaches 87.1%. Notably, the de novo-designed peptide hydrogel selected from this work boosts the immune response of the receptor binding domain of SARS-CoV-2 in the mice model. Our approach taps into the potential of machine learning for predicting peptide hydrogelator and significantly expands the scope of natural peptide hydrogels.

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/s41467-023-39648-2

Cites

62 of 62 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Workflow of coupled experimental and machine learning approach for discovering tetrapeptide hydrogels and their potential biological applications. a 104 uniformly distributed tetrapeptide sequences are obtained by hypercubic sampling first. CGMD simulations are then performed to generate the training data of aggregation propensity (AP), and regression models are trained to predict the AP of the entire sequence space of tetrapeptides (204). b Based on the available score function APH = AP2 × logP0.5, 55 peptides are selected and chemically synthesized to verify the gelation ability (gel marked with 1 and non-gel marked with 0). The sequence (feature) and the 1/0 (label) data are then passed to the ML algorithm to train a classification model, producing a gelation corrector Cg for each tetrapeptide. The APH score is then updated to APHC,1 = AP2 × logP0.5 × Cg,1, and another batch containing 55 peptides is selected based on APHC and is synthesized and verified. Then, the whole 110 sequences (feature) and 1/0 (label) data are employed to update the classification model to generate Cg,2, and the APHC,1 score are updated to APHC,2 = AP2× logP0.5 × Cg,2. Based on APHC,2, the third batch of 55 peptides are selected and chemically synthesized, and Cg,2 and APHC,2 are updated to Cg,3 and APHC,3. c The de novo designed peptide hydrogel is applied to serve as an efficient adjuvant for enhancing antibody production of RBD protein.
Fig. 1 Workflow of coupled experimental and machine learning approach for discovering tetrapeptide hydrogels and their potential biological applications. a 104 uniformly distributed tetrapeptide sequences are obtained by hypercubic sampling first. CGMD simulations are then performed to generate the training data of aggregation propensity (AP), and regression models are trained to predict the AP of the entire sequence space of tetrapeptides (204). b Based on the available score function APH = AP2 × logP0.5, 55 peptides are selected and chemically synthesized to verify the gelation ability (gel marked with 1 and non-gel marked with 0). The sequence (feature) and the 1/0 (label) data are then passed to the ML algorithm to train a classification model, producing a gelation corrector Cg for each tetrapeptide. The APH score is then updated to APHC,1 = AP2 × logP0.5 × Cg,1, and another batch containing 55 peptides is selected based on APHC and is synthesized and verified. Then, the whole 110 sequences (feature) and 1/0 (label) data are employed to update the classification model to generate Cg,2, and the APHC,1 score are updated to APHC,2 = AP2× logP0.5 × Cg,2. Based on APHC,2, the third batch of 55 peptides are selected and chemically synthesized, and Cg,2 and APHC,2 are updated to Cg,3 and APHC,3. c The de novo designed peptide hydrogel is applied to serve as an efficient adjuvant for enhancing antibody production of RBD protein.CC BY · PubMed Central
Fig. 2 “Human-in-the-loop” for obtaining corrected score APHC. a Performance of different algorithms (i.e., LR: linear regression; NN: nearest neighbor; RF: random forest; SVM: support vector machine) with different numbers of training datasets (i.e., 1,000, 5,000, and 10,000). b Training and testing performance of ML model trained with SVM and 10,000 data using one-hot representation. The color scale indicates the density of the data points. c Error distribution with respect to simulated AP value (APsim). d First batch: ranking of experimentally selected peptides with respect to APprd and APH, and accuracy of resulted Cg,1. The Chi-square statistic test (single-sided test) has been performed, with the null hypothesis that the proportion of hydrogel-forming peptides in the population of top 8000 APH score is larger or equal to 61.5%, with a degree of freedom of 1 and significance level of 0.05. e Second batch: ranking of experimentally selected peptides with respect to APH and APHC,1, and accuracy of results in Cg,2. The Chi-square statistic test (single-sided test) has been performed with a degree of freedom of 1 and a significance level of 0.05. f Third batch: ranking of experimentally selected peptides with respect to APH and APHC,2, and accuracy of resulted Cg,3. The Chi-square statistic test (single-sided test) has been performed with a degree of freedom of 1 and a significance level of 0.05. g Gelation hit rate of experimentally synthesized tetrapeptides within the top 8000 ranking with respect to APH (first batch), APHC,1 (second batch), APHC,2 (third batch), and APHC,3 (final). h The comparison between final gelation hit rates evaluated by APprd, APH, and APHC. i (The Chi-square statistic test (single-sided test) has been performed with a degree of freedom of 1 and significance level of 0.05.) and j Distribution of APHC and APH with respect to logP’ of experimentally synthesized 165 tetrapeptides (blue indicates gelation while red indicates solution) and the complete sequence space of tetrapeptides (gray). k Comparison between the ranking of APHC (r-APHC) and APH (r-APH), where color indicates the absolute difference between r-APHC and r-APH of a single tetrapeptide. l TEM images of WPYY, WWCP, WVII, and IMVV (Inserts: optical images of the corresponding peptide). Source data are provided as a Source Data file.
Fig. 2 “Human-in-the-loop” for obtaining corrected score APHC. a Performance of different algorithms (i.e., LR: linear regression; NN: nearest neighbor; RF: random forest; SVM: support vector machine) with different numbers of training datasets (i.e., 1,000, 5,000, and 10,000). b Training and testing performance of ML model trained with SVM and 10,000 data using one-hot representation. The color scale indicates the density of the data points. c Error distribution with respect to simulated AP value (APsim). d First batch: ranking of experimentally selected peptides with respect to APprd and APH, and accuracy of resulted Cg,1. The Chi-square statistic test (single-sided test) has been performed, with the null hypothesis that the proportion of hydrogel-forming peptides in the population of top 8000 APH score is larger or equal to 61.5%, with a degree of freedom of 1 and significance level of 0.05. e Second batch: ranking of experimentally selected peptides with respect to APH and APHC,1, and accuracy of results in Cg,2. The Chi-square statistic test (single-sided test) has been performed with a degree of freedom of 1 and a significance level of 0.05. f Third batch: ranking of experimentally selected peptides with respect to APH and APHC,2, and accuracy of resulted Cg,3. The Chi-square statistic test (single-sided test) has been performed with a degree of freedom of 1 and a significance level of 0.05. g Gelation hit rate of experimentally synthesized tetrapeptides within the top 8000 ranking with respect to APH (first batch), APHC,1 (second batch), APHC,2 (third batch), and APHC,3 (final). h The comparison between final gelation hit rates evaluated by APprd, APH, and APHC. i (The Chi-square statistic test (single-sided test) has been performed with a degree of freedom of 1 and significance level of 0.05.) and j Distribution of APHC and APH with respect to logP’ of experimentally synthesized 165 tetrapeptides (blue indicates gelation while red indicates solution) and the complete sequence space of tetrapeptides (gray). k Comparison between the ranking of APHC (r-APHC) and APH (r-APH), where color indicates the absolute difference between r-APHC and r-APH of a single tetrapeptide. l TEM images of WPYY, WWCP, WVII, and IMVV (Inserts: optical images of the corresponding peptide). Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 3 Experimental investigations on the self-assembly behavior of 165 synthetic tetrapeptides. a TEM images of 6 representative hydrogels of synthetic tetrapeptides, respectively. Inserts: optical images of the corresponding hydrogel (pH between 7.0 to 7.5). MD simulation results (1250 ns) and APHC ranking were shown in the right column. b Dynamic frequency sweep of tetrapeptide hydrogels at the strain value of 0.5%. c FTIR spectra in the amide I region of tetrapeptide hydrogels. d TEM images of 6 representative non-hydrogels of tetrapeptide Insets: optical images of corresponding solution/suspension (pH = 7.5). MD simulation results (1250 ns) and APHC ranking were shown in the right column. e Statistics and classification of morphologies obtained by TEM for hydrogel-forming tetrapeptides (100 peptides). f Statistics and classification of morphologies obtained by TEM for non-hydrogel-forming tetrapeptides (65 peptides). Source data are provided as a Source Data file.
Fig. 3 Experimental investigations on the self-assembly behavior of 165 synthetic tetrapeptides. a TEM images of 6 representative hydrogels of synthetic tetrapeptides, respectively. Inserts: optical images of the corresponding hydrogel (pH between 7.0 to 7.5). MD simulation results (1250 ns) and APHC ranking were shown in the right column. b Dynamic frequency sweep of tetrapeptide hydrogels at the strain value of 0.5%. c FTIR spectra in the amide I region of tetrapeptide hydrogels. d TEM images of 6 representative non-hydrogels of tetrapeptide Insets: optical images of corresponding solution/suspension (pH = 7.5). MD simulation results (1250 ns) and APHC ranking were shown in the right column. e Statistics and classification of morphologies obtained by TEM for hydrogel-forming tetrapeptides (100 peptides). f Statistics and classification of morphologies obtained by TEM for non-hydrogel-forming tetrapeptides (65 peptides). Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 4 Hydrogelation laws from experiments and simulations. a Sequences of 165 synthesized tetrapeptides. The number in each fill represents the rank of APHC. Different color represents the hydrogelation capability at different concentrations. b Averaged rank of APHC of hydrogel-forming peptides at 30, 60, 90, and 120 mM, as well as non-gelating peptides at 120 mM (after the dashed line), n = 46 for 30 mM, n = 30 for 60 mM, n = 24 for 90 and 120 mM (with gelation), and n = 65 for 120 mM (without gelation, n represent the number of synthetic tetrapeptides in each category. c Contribution of each amino acid at different positions to hydrogel formation, compared between 100 experimental data of hydrogel-forming peptides and the top 8,000 APHC simulation data. d Distribution of 8,000 APHC with amino acid F fixed at the C terminus (P4). The x-axis is P1 (N-terminus), the y-axis is P2, and the third position is illustrated in the rectangular box. Source data are provided as a Source Data file.
Fig. 4 Hydrogelation laws from experiments and simulations. a Sequences of 165 synthesized tetrapeptides. The number in each fill represents the rank of APHC. Different color represents the hydrogelation capability at different concentrations. b Averaged rank of APHC of hydrogel-forming peptides at 30, 60, 90, and 120 mM, as well as non-gelating peptides at 120 mM (after the dashed line), n = 46 for 30 mM, n = 30 for 60 mM, n = 24 for 90 and 120 mM (with gelation), and n = 65 for 120 mM (without gelation, n represent the number of synthetic tetrapeptides in each category. c Contribution of each amino acid at different positions to hydrogel formation, compared between 100 experimental data of hydrogel-forming peptides and the top 8,000 APHC simulation data. d Distribution of 8,000 APHC with amino acid F fixed at the C terminus (P4). The x-axis is P1 (N-terminus), the y-axis is P2, and the third position is illustrated in the rectangular box. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 5 Response to tetrapeptide-based hydrogel nano vaccination. a 6–8 weeks C57BL/6 mice were immunized thrice at day 0, 7, and 14 with 15 μg RBD (RBD group), 12.5 μL aluminum adjuvant, and 15 μg RBD (Alum + RBD group), 60 mM tetrapeptide hydrogel and 15 μg RBD (YAWF + RBD group). Serum and splenocytes were collected on day 21. b Enzyme-linked immunosorbent assay (ELISA) responses to serum samples (RBD-specific) at different dilutions. SARS-CoV-2 RBD-specific IgG antibodies (IgG, IgG1, IgG2b, and IgG2c) were analyzed by endpoint dilution ELISA and measured as absorbance at 450 nm. The data were shown as the mean ± SEM (n = 6 biologically independent mice), and the p values were calculated by comparing RBD with YAWF + RBD using a one-way ANOVA test. c 7 days after the last immunization, splenocytes were collected and re-stimulated with RBD protein. The bars shown were mean ± SD (n = 6 biologically independent samples), and differences between RBD and other treatments were determined using a one-way ANOVA test. The secretion of IL-5 and IFN-γ in the splenocytes supernatants was detected using ELISA. d Optical images of bone marrow-derived dendritic cells (BMDCs) treated with RBD-loaded tetrapeptide hydrogel (scale bar = 100 μm). e Flow cytometry analysis of BMDCs expressing CD83, CD80, and CD86. f The level of IL-6 and TNF-α in BMDCs culture supernatants were analyzed using ELISA. The data were shown as the mean ± SD (n = 3 biologically independent samples), and differences between RBD and other treatments were determined using one-way ANOVA test. Source data are provided as a Source Data file.
Fig. 5 Response to tetrapeptide-based hydrogel nano vaccination. a 6–8 weeks C57BL/6 mice were immunized thrice at day 0, 7, and 14 with 15 μg RBD (RBD group), 12.5 μL aluminum adjuvant, and 15 μg RBD (Alum + RBD group), 60 mM tetrapeptide hydrogel and 15 μg RBD (YAWF + RBD group). Serum and splenocytes were collected on day 21. b Enzyme-linked immunosorbent assay (ELISA) responses to serum samples (RBD-specific) at different dilutions. SARS-CoV-2 RBD-specific IgG antibodies (IgG, IgG1, IgG2b, and IgG2c) were analyzed by endpoint dilution ELISA and measured as absorbance at 450 nm. The data were shown as the mean ± SEM (n = 6 biologically independent mice), and the p values were calculated by comparing RBD with YAWF + RBD using a one-way ANOVA test. c 7 days after the last immunization, splenocytes were collected and re-stimulated with RBD protein. The bars shown were mean ± SD (n = 6 biologically independent samples), and differences between RBD and other treatments were determined using a one-way ANOVA test. The secretion of IL-5 and IFN-γ in the splenocytes supernatants was detected using ELISA. d Optical images of bone marrow-derived dendritic cells (BMDCs) treated with RBD-loaded tetrapeptide hydrogel (scale bar = 100 μm). e Flow cytometry analysis of BMDCs expressing CD83, CD80, and CD86. f The level of IL-6 and TNF-α in BMDCs culture supernatants were analyzed using ELISA. The data were shown as the mean ± SD (n = 3 biologically independent samples), and differences between RBD and other treatments were determined using one-way ANOVA test. 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

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

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

Show all 62 references

Research works printed but not linked: 2

  1. Mueller, T., Kusne, A. G. & Ramprasad, R. Machine learning in materials science: Recent progress and emerging applications. Rev. Comput. Chem. 29, 186–273 (2016).
  2. Seber, G. A. & Lee, A. J. Linear regression analysis. (John Wiley & Sons, 2012).

Cited by

Sorted by citations

Items 101–132 of 132

Related

See all

Metadata

Authors
Xu, T., Wang, J., Zhao, S., Chen, D., Zhang, H., Fang, Y., Kong, N., Zhou, Z., Li, W., & Wang, H.
Affiliations
Westlake University (CN)
Corresponding author
  • Li, Wenbin
  • Wang, Huaimin
ORCID
Venue
Nature Communications
Publisher
Springer Science and Business Media LLC
Published
2023-06-30
Volume
14
Issue
1
Pages
3880-3880
ISSN
2041-1723 (ISSN-L)
DOI
10.1038/s41467-023-39648-2
PMC
PMC10313671
OpenAlex
W4382775111
Cited by
132
References
62
Citations by year
2026: 42 · 2025: 44 · 2024: 39 · 2023: 7
FWCI
8.45
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • The HKU Scholars Hub (University of Hong Kong) (submittedVersion)
Funders
National Natural Science Foundation of China (82022038)
Topics
  • Physical Sciences › Materials Science › Biomaterials › Supramolecular Self-Assembly in Materials
  • Physical Sciences › Chemistry › Organic Chemistry › Chemical Synthesis and Analysis
  • Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
Keywords
peptide hydrogels, tetrapeptide, machine learning, human-in-the-loop, interactive machine learning, peptide self-assembly, aggregation propensity, hydrophobicity, de novo peptide design, SARS-CoV-2 receptor-binding domain, immune response
MeSH terms
COVID-19, SARS-CoV-2, Amino Acid Sequence, Animals, Humans, Peptides, Hydrogels, Mice

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