CoChat
Sign Up
PreprintCited by 41Machine Learning in BioinformaticsLipid Membrane Structure and BehaviorProtein Structure and Dynamics

The DOI matches its Crossref record; 1 of the 1 references it cites were found in a registry or an index.

PMIpred: A physics-informed web server for quantitative Protein-Membrane Interaction prediction

Open access
van Hilten, Niek, Verwei, Nino, Methorst, Jeroen, Nase, Carsten, Bernatavicius, Andrius, Risselada, Herre Jelger·2023·Bioinformatics
DOI 10.1101/2023.04.10.536211PMC PMC11212490

Preprint — published as 10.1093/bioinformatics/btae069 · Paper page

Open this paper in CoChat — free account

Abstract

Motivation: Many membrane peripheral proteins have evolved to transiently interact with the surface of (curved) lipid bilayers. Currently, methods to quantitatively predict sensing and binding free energies for protein sequences or structures are lacking, and such tools could greatly benefit the discovery of membrane-interacting motifs, as well as their de novo design. Results: Here, we trained a transformer neural network model on molecular dynamics data for > 50,000 peptides that is able to accurately predict the (relative) membrane-binding free energy for any given amino acid sequence. Using this information, our physics-informed model is able to classify a peptide’s membrane-associative activity as either non-binding, curvature sensing, or membrane binding. Moreover, this method can be applied to detect membraneinteraction regions in a wide variety of proteins, with comparable predictive performance as state-of-the-art data-driven tools like DREAMM, PPM3, and MODA, but with a wider applicability regarding protein diversity, and the added feature to distinguish curvature sensing from general membrane binding. Availability: We made these tools available as a web server, coined Protein-Membrane Interaction predictor (PMIpred), which can be accessed at https://pmipred.fkt.physik.tu-dortmund.de .

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 · 10.1101/2023.04.10.536211

Cites

1 of 1 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1. (A) The membrane-binding probability Pm as a function of the membrane-binding free energy ΔFsm (at R=50 nm) shows a sharp transition. Please note the distinction between ΔFsm and the relative curvature-sensing free energy ΔΔF, which are linearly correlated (Eq. 3 and Fig. 4C in our previous work (van Hilten et al. 2023)). The orange area marks the ‘sensor regime’ (van Hilten et al. [2023]). Insets show cartoon explanations of the three classes. (B) Correlation plot for 96 sequences (all 24 residues long, not part of the training data) spanning the full ΔΔF-range. MD-calculated values are averages and standard deviations for 3 independent replicas (500 ns simulation per run, as described in van Hilten et al. (2023)). (C) Schematic workflow of predicting membrane-interaction regions in protein structures within PMIpred. The trained transformer model predicts ΔΔF (in kJ mol⁻¹) for overlapping peptide segments from the protein sequence. From the 3D protein structure, per-residue SASA values are calculated. Finally, those two features are combined to yield the final quantitative prediction, as visualized by PMIpred. Non-binding and buried residues are shown in purple. Sensing residues are shown in orange. Binding residues are shown in red. The protein structure shown here is the espin1 ENTH domain (PDB: 5onf).
Figure 1. (A) The membrane-binding probability Pm as a function of the membrane-binding free energy ΔFsm (at R=50 nm) shows a sharp transition. Please note the distinction between ΔFsm and the relative curvature-sensing free energy ΔΔF, which are linearly correlated (Eq. 3 and Fig. 4C in our previous work (van Hilten et al. 2023)). The orange area marks the ‘sensor regime’ (van Hilten et al. [2023]). Insets show cartoon explanations of the three classes. (B) Correlation plot for 96 sequences (all 24 residues long, not part of the training data) spanning the full ΔΔF-range. MD-calculated values are averages and standard deviations for 3 independent replicas (500 ns simulation per run, as described in van Hilten et al. (2023)). (C) Schematic workflow of predicting membrane-interaction regions in protein structures within PMIpred. The trained transformer model predicts ΔΔF (in kJ mol⁻¹) for overlapping peptide segments from the protein sequence. From the 3D protein structure, per-residue SASA values are calculated. Finally, those two features are combined to yield the final quantitative prediction, as visualized by PMIpred. Non-binding and buried residues are shown in purple. Sensing residues are shown in orange. Binding residues are shown in red. The protein structure shown here is the espin1 ENTH domain (PDB: 5onf).CC BY · PubMed Central
Figure 2. (A) Users can input a peptide sequence (part of ArfGAP1’s ALPS2 motif (Mesmin et al. 2007), in this example) and chose their desired target membrane. (B) Helical wheel diagram, generated by modlAMP (Müller et al. 2017). (C) Predicted free-energy values and physicochemical characteristics. (D) The position of the input sequence on the non-binding → binding continuum. Predicted sensors fall in the orange transition zone. (E) Membrane-binding probability Pm as a function of vesicle radius R. Again, sensors fall in the orange area.
Figure 2. (A) Users can input a peptide sequence (part of ArfGAP1’s ALPS2 motif (Mesmin et al. 2007), in this example) and chose their desired target membrane. (B) Helical wheel diagram, generated by modlAMP (Müller et al. 2017). (C) Predicted free-energy values and physicochemical characteristics. (D) The position of the input sequence on the non-binding → binding continuum. Predicted sensors fall in the orange transition zone. (E) Membrane-binding probability Pm as a function of vesicle radius R. Again, sensors fall in the orange area.CC BY · PubMed Central
Figure 3. (A–C) Screening results for ArfGAP1 (A, AF: Q8N6T3), the ENTH domain from epsin1 (B, PDB: 5onf), and the N-BAR domain from Bin1 (C, AF: Q9BTH3). Insets show sequences and corresponding predicted classification of putative active regions. Purple: non-binding or SASA below threshold (-). Orange: curvature sensing (S). Red: membrane binding (B). The known membrane-interaction regions are marked in grey. (D) Comparing PMIpred to DREAMM, PPM3, and MODA.
Figure 3. (A–C) Screening results for ArfGAP1 (A, AF: Q8N6T3), the ENTH domain from epsin1 (B, PDB: 5onf), and the N-BAR domain from Bin1 (C, AF: Q9BTH3). Insets show sequences and corresponding predicted classification of putative active regions. Purple: non-binding or SASA below threshold (-). Orange: curvature sensing (S). Red: membrane binding (B). The known membrane-interaction regions are marked in grey. (D) Comparing PMIpred to DREAMM, PPM3, and MODA.CC BY · PubMed Central
Figure 4. (A) The module takes four inputs: (1) The size of the sliding window (from 7 to 24 residues, default = 15). (2) The SASA threshold (default = 0.8 nm2), above which a residue is considered sufficiently solvent-accessible to contribute to membrane interactions. (3) The target membrane. (4) An uploaded PDB-file or PDB ID. (B) Output includes: (5) A downloadable zip-file containing all output files that are also visualized on this page. (6) An interactive visualization of the protein structure using 3Dmol (Rego and Koes 2015). Predicted non-binding, sensing, and binding regions are coloured purple, orange, and red, respectively. (7) An output table containing the full protein sequence and the annotated classification: B, binder; S, sensor; -, non-binder or not accessible (SASA below threshold). (C) A table containing the free energies and classifications of the individual segments that were screened. (D) A more detailed look into the data underlying the table in (B), including the SASA, ΔΔF values, and classification per residue.
Figure 4. (A) The module takes four inputs: (1) The size of the sliding window (from 7 to 24 residues, default = 15). (2) The SASA threshold (default = 0.8 nm2), above which a residue is considered sufficiently solvent-accessible to contribute to membrane interactions. (3) The target membrane. (4) An uploaded PDB-file or PDB ID. (B) Output includes: (5) A downloadable zip-file containing all output files that are also visualized on this page. (6) An interactive visualization of the protein structure using 3Dmol (Rego and Koes 2015). Predicted non-binding, sensing, and binding regions are coloured purple, orange, and red, respectively. (7) An output table containing the full protein sequence and the annotated classification: B, binder; S, sensor; -, non-binder or not accessible (SASA below threshold). (C) A table containing the free energies and classifications of the individual segments that were screened. (D) A more detailed look into the data underlying the table in (B), including the SASA, ΔΔF values, and classification per residue.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

Show all 1 references

Research works printed but not linked: 77

  1. 10.1016/j.bbamem.2016.02.018 · DOI
  2. 10.1016/j.chemphyslip.2015.07.015 · DOI
  3. 10.1038/nature02108 · DOI
  4. 10.1038/nsmb1194 · DOI
  5. 10.1074/jbc.273.16.9443 · DOI
  6. 10.1083/jcb.201011118 · DOI
  7. 10.1016/j.febslet.2009.10.022 · DOI
  8. 2014, Discovery of novel membrane binding structures and functions, Biochem. Cell Biol, 92, 555, 10.1139/bcb-2014-0074 · DOI
  9. 2022, Predicting protein-membrane interfaces of peripheral membrane proteins using ensemble machine learning, Brief. Bioinform, 23, bbab518, 10.1093/bib/bbab518 · DOI · Paper page
  10. 2022, DREAMM: a web-based server for drugging protein-membrane interfaces as a novel workflow for targeted drug design, Bioinformatics, 38, 5449, 10.1093/bioinformatics/btac680 · DOI
  11. 10.1002/pro.4219 · DOI
  12. 2023, Physics-based generative model of curvature sensing peptides; distinguishing sensors from binders, Sci. Adv, 9, eade8839, 10.1126/sciadv.ade8839 · DOI
  13. 2022, Efficient quantification of lipid packing defect sensing by amphipathic peptides: Comparing martini 2 and 3 with CHARMM36, J. Chem. Theory Comput, 18, 4503, 10.1021/acs.jctc.2c00222 · DOI
  14. 2005, An experiment-based algorithm for predicting the partitioning of unfolded peptides into phosphatidylcholine bilayer interfaces, Biochemistry, 44, 12614, 10.1021/bi051193b · DOI
  15. 2017, Polosukhin. Attention is all you need, Adv. Neural Inf. Process. Syst, 30, 6000
  16. 10.1016/0022-2836(73)90011-9 · DOI
  17. 10.1093/bioinformatics/btp163 · DOI · Paper page
  18. 10.1093/bioinformatics/btx285 · DOI · Paper page
  19. 10.1093/bioinformatics/btn392 · DOI
  20. 1983, Hydrophobic parameters-pi of amino-acid side-chains from the partitioning of n-acetyl-amino-acid amides, Eur. J. Med. Chem, 18, 369
  21. 10.1038/299371a0 · DOI
  22. 10.1021/bi062288w · DOI
  23. 10.1126/science.270.5244.1999 · DOI
  24. 10.1093/NAR/GKAB1061 · DOI · Paper page
  25. 10.1126/science.291.5506.1047 · DOI · Paper page
  26. 10.1038/nature01020 · DOI · Paper page
  27. 10.1021/ja907936c · DOI
  28. 10.1074/jbc.M302865200 · DOI
  29. 2006, Membrane topology of helix 0 of the epsin n-terminal homology domain, Mol. Cells, 21, 428, 10.1016/S1016-8478(23)12873-1 · DOI
  30. 10.1016/j.jmb.2012.08.010 · DOI
  31. 10.1529/biophysj.107.113118 · DOI
  32. 10.1529/biophysj.108.134155 · DOI
  33. 10.1126/science.1092586 · DOI · Paper page
  34. 10.1038/srep08026 · DOI
  35. 10.1016/j.tcb.2015.09.005 · DOI
  36. 10.1093/bioinformatics/btu829 · DOI
  37. J. Hallgren , K.D. Tsirigos , M.D. Pedersen , José J. Almagro A., P. Marcatili , H. Nielsen , A. Krogh , and O. Winther . DeepTMHMM predicts alpha and beta transmembrane proteins using deep neural networks. bioRxiv, 2022. · DOI
  38. 10.1093/nar/gkac928 · DOI
  39. T. Akiba , S. Sano , T. Yanase , T. Ohta , and M. Koyama . Optuna: A Next-generation Hyperparameter Optimization Framework, pages 2623–2631. 2019. · DOI
  40. 10.1016/j.bpj.2011.01.036 · DOI
  41. 2021, Martini 3: a general purpose force field for coarse-grained molecular dynamics, Nat. Methods, 18, 382, 10.1038/s41592-021-01098-3 · DOI · Paper page
  42. 10.1021/ct700324x · DOI
  43. 10.1021/cb1001558 · DOI
  44. 10.1073/pnas.1212025109 · DOI · Paper page
  45. 10.1021/bi0477034 · DOI
  46. 2015, Structural basis for the inhibition of voltage-dependent k+ channel by gating modifier toxin, Sci. Rep, 5, 14226, 10.1038/srep14226 · DOI
  47. 10.1038/srep34333 · DOI
  48. 10.1006/jmbi.2000.4283 · DOI
  49. 10.1074/jbc.M601089200 · DOI
  50. 10.1016/j.advenzreg.2007.11.009 · DOI
  51. 2011, Monitoring glycolipid transfer protein activity and membrane interaction with the surface plasmon resonance technique, Biochim. Biophys. Acta, 1808, 47, 10.1016/j.bbamem.2010.08.018 · DOI
  52. 10.1021/acs.biochem.5b00618 · DOI
  53. 10.1074/jbc.M112.367730 · DOI
  54. 10.1091/mbc.E06-01-0089 · DOI
  55. 2000, Use of the parallax-quench method to determine the position of the active-site loop of cholesterol oxidase in lipid bilayers, Biochemistry, 39, 13383, 10.1021/bi001407j · DOI
  56. 10.1021/ja4003525 · DOI
  57. 10.1046/j.1365-313X.2001.01102.x · DOI
  58. 10.1016/j.abb.2013.06.012 · DOI · Paper page
  59. 2017, The role of human monoacylglycerol lipase (hMAGL) binding pocket in breakup of unsaturated phospholipid membranes, Anal. Biochem, 536, 90, 10.1016/j.ab.2017.08.009 · DOI
  60. 10.1046/j.1432-1327.2000.01213.x · DOI
  61. 2018, Lipids shape the electron acceptor-binding site of the peripheral membrane protein dihydroorotate dehydrogenase, Cell Chem. Biol, 25, 309, 10.1016/j.chembiol.2017.12.012 · DOI
  62. 10.1016/S0092-8674(00)81663-3 · DOI
  63. 10.1038/onc.2013.507 · DOI
  64. 1999, A highly active, soluble mutant of the membrane-associated (s)-mandelate dehydrogenase from pseudomonas putida, Biochemistry, 38, 12367, 10.1021/bi990996u · DOI
  65. 2017, Membrane binding of the insertion sequence of proteus vulgaris l-amino acid deaminase stabilizes protein structure and increases catalytic activity, Sci. Rep, 7, 13719, 10.1038/s41598-017-14238-7 · DOI
  66. 10.1021/bi0357356 · DOI
  67. 10.1074/jbc.M511943200 · DOI
  68. 2022, Dissecting peripheral protein-membrane interfaces, PLOS Comput. Biol, 18, e1010346, 10.1371/journal.pcbi.1010346 · DOI · Paper page
  69. 2010, Structural effects of oncogenic PI3Kα mutations, Curr. Top. Microbiol. Immunol, 347, 43
  70. 10.1021/bi001076a · DOI
  71. 10.1038/sj.emboj.7600714 · DOI
  72. 10.1074/jbc.272.2.766 · DOI
  73. P. Papadopoulou , R. van der Pol , N. van Hilten , M.-A. Moradi , M. Ferraz , J. Aerts , H.J. Risselada , G.J.A. Sevink , and A. Kros . Lipase-mediated selective hydrolysis of lipid droplets in phase separated-liposomes. chemRxiv, 2023. · DOI
  74. 10.1073/pnas.1016472108 · DOI · Paper page
  75. 10.1073/pnas.1007974107 · DOI
  76. 10.1074/jbc.M111.271130 · DOI
  77. 2019, Envelope-deforming antiviral peptide derived from influenza virus M2 protein, Biochem. Biophys. Res. Commun, 517, 507, 10.1016/j.bbrc.2019.07.088 · DOI

Metadata

Authors
van Hilten, N., Verwei, N., Methorst, J., Nase, C., Bernatavicius, A., & Risselada, H. J.
Affiliations
  • Leiden University (NL)
  • TU Dortmund University (DE)
Corresponding author
Risselada, Herre Jelger
ORCID
Venue
Bioinformatics
Publisher
openRxiv
Published
2023-04-10
Volume
40
Issue
2
ISSN
1367-4803 (ISSN-L)
DOI
10.1101/2023.04.10.536211
PMC
PMC11212490
OpenAlex
W4391575167
Cited by
41
References
1
Citations by year
2026: 23 · 2025: 8 · 2024: 10
FWCI
5.56
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Bioinformatics (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
  • Deutsche Forschungsgemeinschaft (390677874, EXC2033390677874, EXC2033)
  • Nederlandse Organisatie voor Wetenschappelijk Onderzoek (723.016.005)
  • Gauss Centre for Supercomputing
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Lipid Membrane Structure and Behavior
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
Keywords
protein-membrane interactions, curvature sensing, peptide binding prediction, physics-informed machine learning, transformer models, molecular dynamics, web server
MeSH terms
Amino Acid Sequence, Membrane Proteins, Peptides, Physics, Neural Networks, Computer

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: PMIpred: A physics-informed web server for quantitative Protein-Membrane Interaction prediction. https://app.cochat.ai/papers/10.1101/2023.04.10.536211
The paper
van Hilten, N., Verwei, N., Methorst, J., Nase, C., Bernatavicius, A., & Risselada, H. J. (2023). PMIpred: A physics-informed web server for quantitative Protein-Membrane Interaction prediction. Bioinformatics. https://doi.org/10.1101/2023.04.10.536211