CoChat
Sign Up
Cited by 172Genomics and Chromatin DynamicsChromosomal and Genetic VariationsBioinformatics and Genomic Networks

The DOI matches its Crossref record; 57 of the 57 references it cites were found in a registry or an index.

A supervised learning framework for chromatin loop detection in genome-wide contact maps

Open access
Salameh, Tarik J., Wang, Xiaotao, Song, Fan, Zhang, Bo, Wright, Sage M., Khunsriraksakul, Chachrit, Ruan, Yijun, Yue, Feng·2020·Nature Communications
DOI 10.1038/s41467-020-17239-9PMC PMC7347923

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

Accurately predicting chromatin loops from genome-wide interaction matrices such as Hi-C data is critical to deepening our understanding of proper gene regulation. Current approaches are mainly focused on searching for statistically enriched dots on a genome-wide map. However, given the availability of orthogonal data types such as ChIA-PET, HiChIP, Capture Hi-C, and high-throughput imaging, a supervised learning approach could facilitate the discovery of a comprehensive set of chromatin interactions. Here, we present Peakachu, a Random Forest classification framework that predicts chromatin loops from genome-wide contact maps. We compare Peakachu with current enrichment-based approaches, and find that Peakachu identifies a unique set of short-range interactions. We show that our models perform well in different platforms, across different sequencing depths, and across different species. We apply this framework to predict chromatin loops in 56 Hi-C datasets, and release the results at the 3D Genome Browser.

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/s41467-020-17239-9

Cites

57 of 57 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 A binary classification framework for loop detection in genome-wide contact data. A contact matrix from Hi-C or similar experiment is decomposed into a training set defined by sub-windows either centered at positive interactions from an orthogonal method (ChIA-PET, PLAC-Seq, HiFISH, Capture Hi-C,…) or random loci of similar genomic distance. Hyperparameter tuning within a 3-fold cross-validation is applied to select a random forest model best able to distinguish the two classes. A trained model can then compute per-pixel probabilities in a different contact map from the same platform, with Hi-C depicted here. A greedy pooling algorithm selects the best-scored contacts from clusters of high-probability pixels.
Fig. 1 A binary classification framework for loop detection in genome-wide contact data. A contact matrix from Hi-C or similar experiment is decomposed into a training set defined by sub-windows either centered at positive interactions from an orthogonal method (ChIA-PET, PLAC-Seq, HiFISH, Capture Hi-C,…) or random loci of similar genomic distance. Hyperparameter tuning within a 3-fold cross-validation is applied to select a random forest model best able to distinguish the two classes. A trained model can then compute per-pixel probabilities in a different contact map from the same platform, with Hi-C depicted here. A greedy pooling algorithm selects the best-scored contacts from clusters of high-probability pixels.CC BY · PubMed Central
Fig. 2 Peakachu framework applied in GM12878 Hi-C. a Distance distributions of CTCF ChIA-PET, H3K27ac HiChIP, and HiCCUPS (Hi-C) interactions in GM12878 (top row). Distributions of Peakachu loops predicted from Hi-C after training with CTCF ChIA-PET or H3K27ac HiChIP data, and union of both predictions (bottom row). Interactions in CTCF ChIA-PET were first pooled to remove local redundancy with the same algorithm used by Peakachu. b Proportion of predicted loops with different regulatory element combinations at anchor loci. c CTCF binding patterns and APA analysis of Peakachu predictions. d Overlap of loops predicted by Peakachu models trained with either CTCF ChIA-PET or H3K27ac HiChIP examples, and visualization of interactions predicted from both models. e Different features drive the predictions for CTCF and H3K27ac models. APA plots for loops uniquely predicted by CTCF or H3K27ac models (top row). The feature importance metric from random forests showing which pixels drive the classification most strongly (bottom row). f Fraction and enrichment of Peakachu loop anchors bound for 133 transcription factors and 10 histone modifications. g Fraction of GM12878 interactions in orthogonal experiments recaptured by merged Peakachu loops. h Fraction of Peakachu predictions validated by orthogonal experiments. Source data are available in the Source Data file.
Fig. 2 Peakachu framework applied in GM12878 Hi-C. a Distance distributions of CTCF ChIA-PET, H3K27ac HiChIP, and HiCCUPS (Hi-C) interactions in GM12878 (top row). Distributions of Peakachu loops predicted from Hi-C after training with CTCF ChIA-PET or H3K27ac HiChIP data, and union of both predictions (bottom row). Interactions in CTCF ChIA-PET were first pooled to remove local redundancy with the same algorithm used by Peakachu. b Proportion of predicted loops with different regulatory element combinations at anchor loci. c CTCF binding patterns and APA analysis of Peakachu predictions. d Overlap of loops predicted by Peakachu models trained with either CTCF ChIA-PET or H3K27ac HiChIP examples, and visualization of interactions predicted from both models. e Different features drive the predictions for CTCF and H3K27ac models. APA plots for loops uniquely predicted by CTCF or H3K27ac models (top row). The feature importance metric from random forests showing which pixels drive the classification most strongly (bottom row). f Fraction and enrichment of Peakachu loop anchors bound for 133 transcription factors and 10 histone modifications. g Fraction of GM12878 interactions in orthogonal experiments recaptured by merged Peakachu loops. h Fraction of Peakachu predictions validated by orthogonal experiments. Source data are available in the Source Data file.CC BY · PubMed Central
Fig. 3 Comparison of Peakachu, HiCCUPS, and Fit-Hi-C in GM12878 Hi-C. a Venn diagram of loops predicted by different methods. The original significant interactions from Fit-Hi-C were pooled by the same algorithm used in Peakachu. Parameters of HiCCUPS and Fit-Hi-C were tuned to make the number of loops uniquely predicted by each method similar. b Visualization of loops. c Distance distribution of loops uniquely detected by each method. d CTCF motif patterns of uniquely detected loops by each method. e CTCF binding profile surrounding loop anchors. f Overlap loops uniquely detected by each method with ChIA-PET/HiChIP interactions. Source data are available in the Source Data file.
Fig. 3 Comparison of Peakachu, HiCCUPS, and Fit-Hi-C in GM12878 Hi-C. a Venn diagram of loops predicted by different methods. The original significant interactions from Fit-Hi-C were pooled by the same algorithm used in Peakachu. Parameters of HiCCUPS and Fit-Hi-C were tuned to make the number of loops uniquely predicted by each method similar. b Visualization of loops. c Distance distribution of loops uniquely detected by each method. d CTCF motif patterns of uniquely detected loops by each method. e CTCF binding profile surrounding loop anchors. f Overlap loops uniquely detected by each method with ChIA-PET/HiChIP interactions. Source data are available in the Source Data file.CC BY · PubMed Central
Fig. 4 Detection of loops at lower sequencing depths. a Predicted interactions from down-sampled versions of the GM12878 Hi-C map with levels of validation by orthogonal methods. b Concordance of predictions from datasets with different down-sampled rates. c Visualization of a chromosome 2 locus for Hi-C data and predicted loops using 50% and 1.5% of sequencing reads. d APA profiles and CTCF motif orientations of 50% and 1.5% loop predictions. The color bar indicates contact counts normalized by the mean of pixels within their windows. Source data are available in the Source Data file.
Fig. 4 Detection of loops at lower sequencing depths. a Predicted interactions from down-sampled versions of the GM12878 Hi-C map with levels of validation by orthogonal methods. b Concordance of predictions from datasets with different down-sampled rates. c Visualization of a chromosome 2 locus for Hi-C data and predicted loops using 50% and 1.5% of sequencing reads. d APA profiles and CTCF motif orientations of 50% and 1.5% loop predictions. The color bar indicates contact counts normalized by the mean of pixels within their windows. Source data are available in the Source Data file.CC BY · PubMed Central
Fig. 5 Application of Peakachu in other Hi-C datasets of different tissues and species. a Overlap of loops predicted in K562 Hi-C by Peakachu models trained in either GM12878 or K562 data. Both models were trained with CTCF ChIA-PET examples. b CTCF binding orientations for predicted loops. c Regulatory elements at anchor loci of predicted loops. d–f Repeat analysis for loops predicted in mESC Hi-C by Peakachu models trained with either GM12878 CTCF ChIA-PET or mESC SMC1 HiChIP. Source data are available in the Source Data file.
Fig. 5 Application of Peakachu in other Hi-C datasets of different tissues and species. a Overlap of loops predicted in K562 Hi-C by Peakachu models trained in either GM12878 or K562 data. Both models were trained with CTCF ChIA-PET examples. b CTCF binding orientations for predicted loops. c Regulatory elements at anchor loci of predicted loops. d–f Repeat analysis for loops predicted in mESC Hi-C by Peakachu models trained with either GM12878 CTCF ChIA-PET or mESC SMC1 HiChIP. Source data are available in the Source Data file.CC BY · PubMed Central
Fig. 6 Cross-platform comparison of Peakachu loops in different tissue types. a Overlap of loops predicted in H1ESC by Micro-C and Hi-C, using the same CTCF ChIA-PET training set. b Patterns of CTCF binding site orientations in predicted loops. c Regulatory elements at anchor loci of predicted loops. d–f Repeat analysis comparing DNA SPRITE with Hi-C in GM12878, using the same CTCF ChIA-PET training set. Source data are available in the Source Data file.
Fig. 6 Cross-platform comparison of Peakachu loops in different tissue types. a Overlap of loops predicted in H1ESC by Micro-C and Hi-C, using the same CTCF ChIA-PET training set. b Patterns of CTCF binding site orientations in predicted loops. c Regulatory elements at anchor loci of predicted loops. d–f Repeat analysis comparing DNA SPRITE with Hi-C in GM12878, using the same CTCF ChIA-PET training set. Source data are available in the 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 57 references

Cited by

Sorted by citations

Items 101–172 of 172

Related

See all

Metadata

Authors
Salameh, T. J., Wang, X., Song, F., Zhang, B., Wright, S. M., Khunsriraksakul, C., Ruan, Y., & Yue, F.
Affiliations
  • Pennsylvania State University (US)
  • Northwestern University (US)
  • The Jackson Laboratory for Genomic Medicine (US)
Corresponding author
  • Wang, Xiaotao
  • Yue, Feng
ORCID
Venue
Nature Communications
Publisher
Springer Science and Business Media LLC
Published
2020-07-09
Volume
11
Issue
1
Pages
3428-3428
ISSN
2041-1723 (ISSN-L)
DOI
10.1038/s41467-020-17239-9
PMC
PMC7347923
OpenAlex
W3041895828
Cited by
165
References
57
Citations by year
2026: 26 · 2025: 33 · 2024: 34 · 2023: 32 · 2022: 18 · 2021: 18 · 2020: 4
FWCI
6.33
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion)
  • 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 Mouseion at the JAXlibrary (Jackson Laboratory) (submittedVersion)
Funders
  • U.S. Department of Health & Human Services | NIH | National Human Genome Research Institute (R01HG009906)
  • U.S. Department of Health & Human Services | NIH | National Institute of General Medical Sciences (R35GM124820)
  • U.S. Department of Health & Human Services | NIH | National Institute of Diabetes and Digestive and Kidney Diseases (R24DK106766)
  • U.S. Department of Health & Human Services | NIH | NCI | Division of Cancer Epidemiology and Genetics, National Cancer Institute (U01CA200060)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Chromatin Dynamics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Chromosomal and Genetic Variations
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
Keywords
Hi-C contact maps, random forest classification, 3D genome organization, gene expression regulation, ChIA-PET, HiChIP, Capture Hi-C
MeSH terms
Supervised Machine Learning, Chromatin, Humans, Organ Specificity, Species Specificity, Genome, Sequence Analysis, DNA, K562 Cells, Databases, Genetic

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