CoChat

Multicoil2: Predicting Coiled Coils and Their Oligomerization States from Sequence in the Twilight Zone

Open access

The DOI matches its Crossref record; 42 of the 44 references it cites were found in a registry or an index; 1 of the 14 sources its claims cite were checked for support.

Abstract

The alpha-helical coiled coil can adopt a variety of topologies, among the most common of which are parallel and antiparallel dimers and trimers. We present Multicoil2, an algorithm that predicts both the location and oligomerization state (two versus three helices) of coiled coils in protein sequences.

Overview

CoChat Summary

TL;DR

Multicoil2 is a Markov Random Field-based method for predicting both coiled-coil regions and their dimer-versus-trimer oligomerization state from sequence. Using a new coiled-coil database, it improves over retrained Multicoil and Paircoil2, especially in leave-family-out tests meant to mimic low-similarity (“twilight zone”) cases.

Key findings

  • The new NPS coiled-coil database contains 2,105 sequences and 124,088 residues, including 1,279 dimers and 333 trimers; the paper notes only dimer, trimer, and tetramer annotations were represented, and no other oligomerization states.
  • In stringent leave-family-out cross-validation, retrained Multicoil at default settings recognized 64.3% of trimer sequences and 88.1% of dimer sequences, whereas Multicoil2 detected 82.5% of trimer sequences at a similar dimer recognition level of 87.2%.
  • For residue-by-residue coiled-coil detection, Multicoil2 achieved 0.30% false positives and 91.8% detection rate under the stringent leave-family-out protocol.
  • The paper reports that Multicoil2 significantly outperforms retrained Multicoil and Paircoil2 for both coiled-coil detection and oligomerization-state prediction, with especially strong performance in the “twilight zone” of sequence identity.

Methods

  • The authors constructed the NPS database from three sources: the Paircoil2 training set, coiled coils detected in the PDB using SOCKET, and new coiled-coil families from the literature; sequences were filtered for quality, heptad consistency, length, and ≤90% sequence identity.
  • Multicoil2 models coiled coils as a Markov Random Field over hidden heptad states and observed amino acids, with multinomial logistic regression used to optimize scoring based on 8 sequence features.
  • Performance was evaluated with leave-family-out cross-validation, training on all but one family plus negative databases, and then testing on the held-out family; coiled-coil detection was assessed against a non-coiled-coil database (PDB-minus / PQS-minus), and oligomerization was evaluated by comparing dimer vs trimer scores.

Limitations

  • The paper states that performance varies among coiled-coil families under stringent leave-family-out testing, and that it is difficult to determine why some families do better or worse.
  • The authors note a bias toward predicting coiled coils in the model that required an arbitrary correction (subtracting 20 from dimer and trimer constants), which suggests some calibration fragility.
  • Evident from the text: the database covers only dimers, trimers, and tetramers, and the method is framed around dimer-vs-trimer prediction, so other oligomerization states are not directly addressed.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1371/journal.pone.0023519

Cites

42 of 42 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Flow-chart overview of the Multicoil2 method. From the training set of labelled dimer, trimer and negative coiled coil sequences, we compute the probability of each amino acid at each heptad position in dimer sequences and trimer sequences. Also, we compute the probability of each amino acid in negative sequences. From the resulting frequency tables shown in the upper left, along with the training sequences shown in the upper right, we compute sequence features for each training sequence. Running a multinomial logistic regression on these values generates a three-way classifier which is used in the MRF as described in Methods.
Figure 1 Flow-chart overview of the Multicoil2 method. From the training set of labelled dimer, trimer and negative coiled coil sequences, we compute the probability of each amino acid at each heptad position in dimer sequences and trimer sequences. Also, we compute the probability of each amino acid in negative sequences. From the resulting frequency tables shown in the upper left, along with the training sequences shown in the upper right, we compute sequence features for each training sequence. Running a multinomial logistic regression on these values generates a three-way classifier which is used in the MRF as described in Methods.CC BY · PubMed Central
Figure 2 Dimer versus trimer recognition. Multicoil2 and Multicoil ROC curves based on leave-family-out cross validation for per-residue (a) and per-sequence (b) recognition.
Figure 2 Dimer versus trimer recognition. Multicoil2 and Multicoil ROC curves based on leave-family-out cross validation for per-residue (a) and per-sequence (b) recognition.CC BY · PubMed Central
Figure 3 Coiled coil detection. Multicoil2, retrained Multicoil and Paircoil2 ROC curves based on leave-family-out cross validation for per-residue (a) and per-sequence (b) detection. Note the very small values on x-axis for false positive rate.
Figure 3 Coiled coil detection. Multicoil2, retrained Multicoil and Paircoil2 ROC curves based on leave-family-out cross validation for per-residue (a) and per-sequence (b) detection. Note the very small values on x-axis for false positive rate.CC BY · PubMed Central
Figure 4 Overview of training of the Multicoil and Multicoil2 algorithms. (a) The raw scores used to generate the Multicoil gaussians for each of n-1 families are computed based on frequency tables generated from the other n-2 families. (b) The Multicoil2 sequence features for each of n-1 families are computed based on frequency tables generated from the other n-2 families. Those features are used to find the regression coefficients, which determine the MRF.
Figure 4 Overview of training of the Multicoil and Multicoil2 algorithms. (a) The raw scores used to generate the Multicoil gaussians for each of n-1 families are computed based on frequency tables generated from the other n-2 families. (b) The Multicoil2 sequence features for each of n-1 families are computed based on frequency tables generated from the other n-2 families. Those features are used to find the regression coefficients, which determine the MRF.CC BY · PubMed Central
Figure 5 Graph of allowed transitions. There are two copies of each state, one for the dimer state and one for the trimer state (the transitions of the trimer states are omitted - they are identical to the dimer transitions). The “0” nodes at the top and bottom of the figure refer to the same non-coiled-coil state.
Figure 5 Graph of allowed transitions. There are two copies of each state, one for the dimer state and one for the trimer state (the transitions of the trimer states are omitted - they are identical to the dimer transitions). The “0” nodes at the top and bottom of the figure refer to the same non-coiled-coil state.CC BY · PubMed Central
Figure 6 Example of hidden states corresponding to a given amino acid sequence in the Multicoil2 model.
Figure 6 Example of hidden states corresponding to a given amino acid sequence in the Multicoil2 model.CC BY · PubMed Central

Verification

Claims the paper makes

1 Valid · 1 Not valid · 9 Not checked · 19 Own results and interpretations

8 more claims

Create a free account to see all 30 claims and to verify the 5 references still open.

See all claims — free account
  • It may state two separate claims as one. 15
  • Probably not a substantial claim. 6
  • The quote only partly states this claim. 1

Create a free account to see these claims.

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?

Citations

Cites

Show all 42 references

Cited by

See all 91

Metadata

Authors
Trigg, J., Gutwin, K., Keating, A. E., & Berger, B.
Affiliations
  • MIT Computer Science and Artificial Intelligence Laboratory (US)
  • Massachusetts Institute of Technology (US)
Corresponding author
  • Keating, Amy E.
  • Berger, Bonnie
Venue
PLoS One
Published
2011-08-25
Volume
6
Issue
8
Pages
e23519
ISSN
1932-6203 (ISSN-L)
OpenAlex
W1982005006
Cited by
91
References
44
Citations by year
2026: 2 · 2025: 7 · 2024: 7 · 2023: 7 · 2022: 8 · 2021: 8 · 2020: 5 · 2019: 5 · 2018: 3 · 2017: 7 · 2016: 6 · 2015: 4 · 2014: 8 · 2013: 4 · 2012: 9
FWCI
2.72
Type
journal article
Language
en
Open access
Locations
  • PLoS ONE (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (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)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc)
Funders
  • National Science Foundation (0347203, 0821391)
  • National Institute of General Medical Sciences (R01GM081871)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
Keywords
alpha-helical coiled coil, Markov random fields, hidden Markov models, multinomial logistic regression, protein structure prediction
MeSH terms
Algorithms, Proteins, Software, Protein Structure, Secondary, Protein Multimerization

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.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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: Multicoil2: Predicting Coiled Coils and Their Oligomerization States from Sequence in the Twilight Zone. https://app.cochat.ai/papers/10.1371/journal.pone.0023519
The paper
Trigg, J., Gutwin, K., Keating, A. E., & Berger, B. (2011). Multicoil2: Predicting Coiled Coils and Their Oligomerization States from Sequence in the Twilight Zone. PLoS One. https://doi.org/10.1371/journal.pone.0023519

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.