CoChat
Sign Up
Cited by 268Gene Regulatory Network AnalysisBioinformatics and Genomic NetworksFungal and yeast genetics research

The DOI matches its Crossref record; 64 of the 64 references it cites were found in a registry or an index.

Passing Messages between Biological Networks to Refine Predicted Interactions

Open access
Glass, Kimberly, Huttenhower, Curtis, Quackenbush, John, Yuan, Guo-Cheng·2013·PLoS ONE
DOI 10.1371/journal.pone.0064832PMC PMC3669401
Open this paper in CoChat — free accountOpen PDF

Abstract

Regulatory network reconstruction is a fundamental problem in computational biology. There are significant limitations to such reconstruction using individual datasets, and increasingly people attempt to construct networks using multiple, independent datasets obtained from complementary sources, but methods for this integration are lacking. We developed PANDA (Passing Attributes between Networks for Data Assimilation), a message-passing model using multiple sources of information to predict regulatory relationships, and used it to integrate protein-protein interaction, gene expression, and sequence motif data to reconstruct genome-wide, condition-specific regulatory networks in yeast as a model. The resulting networks were not only more accurate than those produced using individual data sets and other existing methods, but they also captured information regarding specific biological mechanisms and pathways that were missed using other methodologies. PANDA is scalable to higher eukaryotes, applicable to specific tissue or cell type data and conceptually generalizable to include a variety of regulatory, interaction, expression, and other genome-scale data. An implementation of the PANDA algorithm is available at www.sourceforge.net/projects/panda-net.

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 · 10.1371/journal.pone.0064832

Cites

64 of 64 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Outline of the PANDA approach for regulatory network inference integrating three data types. (A) A conceptual illustration showing the generalized framework for the message-passing procedure. (B) An illustration of how the message-passing procedure is applied in assimilating data that represents several various components of biological regulation. The networks are initialized from sequence motif data, physical protein interactions, and co-expression, respectively. The method iteratively passes messages within and among networks to emphasize agreement regarding the TF-gene regulatory relationships occurring within a system. At each time step regulatory (W), co-regulatory (C), and protein-cooperativity (P) networks are updated by passing information between the regulatory network, that reflects potential paths for regulation in the biological system, and the data-specific networks, that reflect “static” pair-wise information shared between gene products and TFs. At convergence, the method provides harmonized expression and interaction modules specific to a biological condition of interest, as well as the output regulatory network controlling those modules in each condition.
Figure 1 Outline of the PANDA approach for regulatory network inference integrating three data types. (A) A conceptual illustration showing the generalized framework for the message-passing procedure. (B) An illustration of how the message-passing procedure is applied in assimilating data that represents several various components of biological regulation. The networks are initialized from sequence motif data, physical protein interactions, and co-expression, respectively. The method iteratively passes messages within and among networks to emphasize agreement regarding the TF-gene regulatory relationships occurring within a system. At each time step regulatory (W), co-regulatory (C), and protein-cooperativity (P) networks are updated by passing information between the regulatory network, that reflects potential paths for regulation in the biological system, and the data-specific networks, that reflect “static” pair-wise information shared between gene products and TFs. At convergence, the method provides harmonized expression and interaction modules specific to a biological condition of interest, as well as the output regulatory network controlling those modules in each condition.CC BY · PubMed Central
Figure 2 An evaluation of PANDA’s performance. (A) The significance of the area under the ROC curve (AUC) for the regulatory network predicted by PANDA at each step during convergence using all experimental data types. Red dotted and dashed lines indicate the AUC values for either random (0.5) or the motif prior (0.687), respectively. Edges included in the motif prior as well as those that are not included in the motif prior are evaluated separately. As messages are passed, the quality of the regulatory network increases. A large portion of this improvement is attributable to a removal of false positives from motif-edges. (B) PANDA’s performance as noise is added to the motif prior. Even upon “full” randomization of the initial motif network, PANDA is able to improve the network prediction, indicating that it can still find biological signal in the absence of an accurate prior. (C) Evaluation of the accuracy of transformation for each data type specific network by PANDA. The initial AUC of each input network is shown as well as the AUC of the edges predicted by PANDA. The significance was determined by jackknifing the input data.
Figure 2 An evaluation of PANDA’s performance. (A) The significance of the area under the ROC curve (AUC) for the regulatory network predicted by PANDA at each step during convergence using all experimental data types. Red dotted and dashed lines indicate the AUC values for either random (0.5) or the motif prior (0.687), respectively. Edges included in the motif prior as well as those that are not included in the motif prior are evaluated separately. As messages are passed, the quality of the regulatory network increases. A large portion of this improvement is attributable to a removal of false positives from motif-edges. (B) PANDA’s performance as noise is added to the motif prior. Even upon “full” randomization of the initial motif network, PANDA is able to improve the network prediction, indicating that it can still find biological signal in the absence of an accurate prior. (C) Evaluation of the accuracy of transformation for each data type specific network by PANDA. The initial AUC of each input network is shown as well as the AUC of the edges predicted by PANDA. The significance was determined by jackknifing the input data.CC BY · PubMed Central
Figure 3 A quantitative and qualitative comparison of the networks predicted by PANDA to those predicted by other network reconstruction algorithms. (A) The AUC of the networks predicted by PANDA, SEREND [50], ReMoDiscovery [11], CLR [2] and C3Net [4] using three distinct sets of expression conditions as an input: regulator knock-out (cyan), cell-cycle (magenta) and stress-response (yellow). PANDA, SEREND and ReMoDiscovery all take prior regulatory information from motif data as an input (indicated with an asterisk) whereas CLR and C3Net derive networks using only expression data. Red dotted and dashed lines indicate the AUC values for either random (0.5) or the motif prior (0.687), respectively. (B) The sensitivity and precision at 90% specificity reported for the condition-specific networks predicted by PANDA, SEREND and CLR. By this measure, PANDA outperforms both SEREND and CLR. (C–E) The overlap of the top 1000 edges by weight from the networks predicted by (C) PANDA, (D) SEREND, or (E) CLR using three distinct sources of gene expression. Many edges are common between the networks predicted by PANDA, but there are also subnetworks of edges unique to each data type that may highlight distinct regulatory programs. In contrast to PANDA, fewer edges are common between the networks predicted by SEREND and almost none are in common between the networks predicted by CLR. (F–G) Functional analysis of genes belonging to each of the condition-specific subnetworks identified with (F) PANDA or (G) CLR. GO categories enriched at Benjamini-Hochberg FDR less than 10−5 and which contain at least 10% of the members in one of the condition-specific gene sets are shown. No categories were enriched at this level for genes belong to the condition-specific subnetworks identified by SEREND.
Figure 3 A quantitative and qualitative comparison of the networks predicted by PANDA to those predicted by other network reconstruction algorithms. (A) The AUC of the networks predicted by PANDA, SEREND [50], ReMoDiscovery [11], CLR [2] and C3Net [4] using three distinct sets of expression conditions as an input: regulator knock-out (cyan), cell-cycle (magenta) and stress-response (yellow). PANDA, SEREND and ReMoDiscovery all take prior regulatory information from motif data as an input (indicated with an asterisk) whereas CLR and C3Net derive networks using only expression data. Red dotted and dashed lines indicate the AUC values for either random (0.5) or the motif prior (0.687), respectively. (B) The sensitivity and precision at 90% specificity reported for the condition-specific networks predicted by PANDA, SEREND and CLR. By this measure, PANDA outperforms both SEREND and CLR. (C–E) The overlap of the top 1000 edges by weight from the networks predicted by (C) PANDA, (D) SEREND, or (E) CLR using three distinct sources of gene expression. Many edges are common between the networks predicted by PANDA, but there are also subnetworks of edges unique to each data type that may highlight distinct regulatory programs. In contrast to PANDA, fewer edges are common between the networks predicted by SEREND and almost none are in common between the networks predicted by CLR. (F–G) Functional analysis of genes belonging to each of the condition-specific subnetworks identified with (F) PANDA or (G) CLR. GO categories enriched at Benjamini-Hochberg FDR less than 10−5 and which contain at least 10% of the members in one of the condition-specific gene sets are shown. No categories were enriched at this level for genes belong to the condition-specific subnetworks identified by SEREND.CC BY · PubMed Central
Figure 4 Specific examples of condition-specific genes and edges highlighted by PANDA. (A) A table of genes (transcription factors bolded) with an enrichment of edges in a particular condition-specific network compared the union of all the networks and an example of their functional role in the cell. Also, the expression levels of these genes across the conditions in the cell-cycle and stress-response datasets. For visualization purposes, each row in each dataset was normalized to a Z-score. The co-expression of the genes in these regulatory “modules” is easily discernible. As co-expression between genes and transcription factors is not used by PANDA when building the networks it is not surprising that some of the TFs are not as highly co-expressed with the other identified genes. (B–C) Visualization of the edges surrounding these enriched genes in the (B) cell cycle and (C) stress response condition-specific networks. Co-regulatory (C) and protein-cooperativity (P) network edges are shown if they are in the top 10% of edges identified by PANDA in the final condition-specific co-regulatory and protein-cooperativity networks (for more information see Methods S1).
Figure 4 Specific examples of condition-specific genes and edges highlighted by PANDA. (A) A table of genes (transcription factors bolded) with an enrichment of edges in a particular condition-specific network compared the union of all the networks and an example of their functional role in the cell. Also, the expression levels of these genes across the conditions in the cell-cycle and stress-response datasets. For visualization purposes, each row in each dataset was normalized to a Z-score. The co-expression of the genes in these regulatory “modules” is easily discernible. As co-expression between genes and transcription factors is not used by PANDA when building the networks it is not surprising that some of the TFs are not as highly co-expressed with the other identified genes. (B–C) Visualization of the edges surrounding these enriched genes in the (B) cell cycle and (C) stress response condition-specific networks. Co-regulatory (C) and protein-cooperativity (P) network edges are shown if they are in the top 10% of edges identified by PANDA in the final condition-specific co-regulatory and protein-cooperativity networks (for more information see Methods S1).CC BY · PubMed Central
Figure 5 The top edges predicted by PANDA, excluding those common among all three predicted networks. Several key TFs have been identified and labeled, including the cell cycle regulator MBP1 and the stress response factor MSN4.
Figure 5 The top edges predicted by PANDA, excluding those common among all three predicted networks. Several key TFs have been identified and labeled, including the cell cycle regulator MBP1 and the stress response factor MSN4.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 64 references

Cited by

Sorted by citations

See all 268

Related

Metadata

Authors
Glass, K., Huttenhower, C., Quackenbush, J., & Yuan, G.-C.
Affiliations
Harvard University (US)
Corresponding author
Yuan, Guo-Cheng
ORCID
Venue
PLoS ONE
Publisher
Public Library of Science (PLoS)
Published
2013-05-31
Volume
8
Issue
5
Pages
e64832
ISSN
1932-6203 (ISSN-L)
DOI
10.1371/journal.pone.0064832
PMC
PMC3669401
OpenAlex
W2020672230
Cited by
266
References
64
Citations by year
2026: 24 · 2025: 26 · 2024: 56 · 2023: 22 · 2022: 24 · 2021: 29 · 2020: 24 · 2019: 15 · 2018: 11 · 2017: 12 · 2016: 12 · 2015: 7 · 2014: 4
FWCI
3.44
Type
journal article
Language
en
Open access
Licence
Locations
  • PLoS ONE (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, other-oa)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Digital Access to Scholarship at Harvard (DASH) (Harvard University) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Figshare (submittedVersion, cc-by)
Funders
  • National Science Foundation (1053486)
  • National Institutes of Health (5r01hl111759-04)
  • National Heart, Lung, and Blood Institute (R01-HL111759, P01HL105339)
  • Division of Biological Infrastructure (DBI-1053486)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Fungal and yeast genetics research
Keywords
regulatory network reconstruction, data integration, message passing, protein-protein interaction, gene expression, sequence motifs, yeast, genome-scale data, network prediction
MeSH terms
Algorithms, Computer Simulation, Models, Biological, Saccharomyces cerevisiae, Reproducibility of Results, Signal Transduction, Gene Regulatory Networks, Protein Interaction Maps

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