CoChat

Towards a Rigorous Assessment of Systems Biology Models: The DREAM3 Challenges

Open access
Prill, Robert J., Marbach, Daniel, Saez-Rodriguez, Julio, Sorger, Peter K., Alexopoulos, Leonidas G., Xue, Xiaowei, Clarke, Neil D., Altan-Bonnet, Gregoire, Stolovitzky, Gustavo · 2010 · PLoS ONE

The DOI matches its Crossref record; 33 of the 33 references it cites were found in a registry or an index.

Abstract

BACKGROUND: Systems biology has embraced computational modeling in response to the quantitative nature and increasing scale of contemporary data sets. The onslaught of data is accelerating as molecular profiling technology evolves. The Dialogue for Reverse Engineering Assessments and Methods (DREAM) is a community effort to catalyze discussion about the design, application, and assessment of systems biology models through annual reverse-engineering challenges. METHODOLOGY AND PRINCIPAL FINDINGS: We describe our assessments of the four challenges associated with the third DREAM conference which came to be known as the DREAM3 challenges: signaling cascade identification, signaling response prediction, gene expression prediction, and the DREAM3 in silico network challenge. The challenges, based on anonymized data sets, tested participants in network inference and prediction of measurements. Forty teams submitted 413 predicted networks and measurement test sets. Overall, a handful of best-performer teams were identified, while a majority of teams made predictions that were equivalent to random. Counterintuitively, combining the predictions of multiple teams (including the weaker teams) can in some cases improve predictive power beyond that of any single method. CONCLUSIONS: DREAM provides valuable feedback to practitioners of systems biology modeling. Lessons learned from the predictions of the community provide much-needed context for interpreting claims of efficacy of algorithms described in the scientific literature.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1371/journal.pone.0009202

Cites

33 of 33 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 The objective of the signaling cascade identification challenge was to identify some of the molecular species in this diagram from single-cell flow cytometry measurements. The upstream binding of a ligand to a receptor and the downstream phosphorylation of a protein are illustrated.
Figure 1 The objective of the signaling cascade identification challenge was to identify some of the molecular species in this diagram from single-cell flow cytometry measurements. The upstream binding of a ligand to a receptor and the downstream phosphorylation of a protein are illustrated.CC BY · PubMed Central
Figure 2 The objective of the signaling response prediction challenge was to predict the concentrations of phosphoproteins and cytokines in response to combinatorial perturbations the environmental cues (stimuli) and perturbations of the signaling network (inhibtors). (a) A compendium of phosphoprotein and cytokine measurements was provided as a training set. (b) Histograms (log scale) of the scoring metric (normalized squared error) for 100,000 random predictions were approximately Gaussian (fitted blue points). Significance of the predictions of the teams (black points) was assessed with respect to the empirical probability densities embodied by these histograms. Scores of the best-performer teams are denoted with arrows.
Figure 2 The objective of the signaling response prediction challenge was to predict the concentrations of phosphoproteins and cytokines in response to combinatorial perturbations the environmental cues (stimuli) and perturbations of the signaling network (inhibtors). (a) A compendium of phosphoprotein and cytokine measurements was provided as a training set. (b) Histograms (log scale) of the scoring metric (normalized squared error) for 100,000 random predictions were approximately Gaussian (fitted blue points). Significance of the predictions of the teams (black points) was assessed with respect to the empirical probability densities embodied by these histograms. Scores of the best-performer teams are denoted with arrows.CC BY · PubMed Central
Figure 3 Overlay of the assignment tables from the seven teams in the signaling cascade identification challenge. The number of teams making each assignment and the p-value is indicated. The p-value expresses the probability of a such a concentration of random guesses in the same table entry. Highlighted entries are correct. Five teams correctly identified species x1 as the kinase, a significant event for the community despite that no team had a significant individual performance.
Figure 3 Overlay of the assignment tables from the seven teams in the signaling cascade identification challenge. The number of teams making each assignment and the p-value is indicated. The p-value expresses the probability of a such a concentration of random guesses in the same table entry. Highlighted entries are correct. Five teams correctly identified species x1 as the kinase, a significant event for the community despite that no team had a significant individual performance.CC BY · PubMed Central
Figure 4 The objective of the gene expression prediction challenge was to predict temporal expression of 50 genes that were withheld from a training set consisting of 9285 genes. (a) Clustered heatmaps of the predicted genes (columns) reveal that two best-performer teams predicted substantially similar gene expression values, though different methods were employed. Results for the 60 minute time-point are shown. (b) The benefits of combining the predictions of multiple teams into a consensus prediction are illustrated by the rank sum prediction (triangles). Some rank sum predictions score higher than the best-performer, depending on the teams that are included. The highest score is achieved by a combination of the predictions of the best four teams.
Figure 4 The objective of the gene expression prediction challenge was to predict temporal expression of 50 genes that were withheld from a training set consisting of 9285 genes. (a) Clustered heatmaps of the predicted genes (columns) reveal that two best-performer teams predicted substantially similar gene expression values, though different methods were employed. Results for the 60 minute time-point are shown. (b) The benefits of combining the predictions of multiple teams into a consensus prediction are illustrated by the rank sum prediction (triangles). Some rank sum predictions score higher than the best-performer, depending on the teams that are included. The highest score is achieved by a combination of the predictions of the best four teams.CC BY · PubMed Central
Figure 5 The objective of the in silico network inference challenge was to infer networks of various sizes (10, 50, and 100 nodes) from steady-state and time-series “measurements” of simulated gene regulation networks. Predicted networks were evaluated on the basis of two scoring metrics, (a) area under the ROC curve and (b) area under the precision-recall curve. ROC and precision-recall curves of the five best teams in the 100-node sub-challenge. (a) Dotted diagonal line is the expected value of a random prediction. (b) Note that the best and second-best performers have different precision-recall characteristics. (c) Histograms (log scale) of the AUROC scoring metric for 100,000 random predictions was approximately Gaussian (fitted blue points) whereas the histogram of the AUPR metric was not (inset). Significance of the predictions of the teams (black points) was assessed with respect to the empirical probability densities embodied by these histograms. Scores of the best-performer team are denoted with arrows. All plots are analyses of the gold standard network called InSilico_Size100_Yeast2.
Figure 5 The objective of the in silico network inference challenge was to infer networks of various sizes (10, 50, and 100 nodes) from steady-state and time-series “measurements” of simulated gene regulation networks. Predicted networks were evaluated on the basis of two scoring metrics, (a) area under the ROC curve and (b) area under the precision-recall curve. ROC and precision-recall curves of the five best teams in the 100-node sub-challenge. (a) Dotted diagonal line is the expected value of a random prediction. (b) Note that the best and second-best performers have different precision-recall characteristics. (c) Histograms (log scale) of the AUROC scoring metric for 100,000 random predictions was approximately Gaussian (fitted blue points) whereas the histogram of the AUPR metric was not (inset). Significance of the predictions of the teams (black points) was assessed with respect to the empirical probability densities embodied by these histograms. Scores of the best-performer team are denoted with arrows. All plots are analyses of the gold standard network called InSilico_Size100_Yeast2.CC BY · PubMed Central
Figure 6 Analysis of the community of teams reveals characteristics of identifiable and unidentifiable network edges. The number of teams that identify an edge at a specified cutoff is a measure of how easy or difficult an edge is to identify. In this analysis we use a cutoff of 2P (i.e., twice the number of actual positive edges in the gold standard network). (a) Histograms indicate the number of teams that correctly identified the edges of the gold standard network called InSilico_Size100_Ecoli1. The ten worst teams in the 100-node sub-challenge identified about the same number of edges as is expected by chance. By contrast, the ten best teams identified more edges than is expected by chance and this sub-community has a markedly different identifiability distribution than random. Still, some edges were not identified by even the ten best teams (see bin corresponding to zero teams). Unidentified edges are characterized by (b) a property of the measurement data and (c) a topological property of the network. (b) Unidentified edges have a lower null-mutant absolute z-score than those that were identified by at least one of the ten best teams. This metric is a measure of the information content of the measurements. (c) Unidentified edges belong to target nodes with a higher in-degree than edges that were identified by at least one of the ten best teams. Circles denote the median and bars denote upper and lower quartiles. Statistics were not computed for bins containing less than four edges. (d) The benefits of combining the predictions of multiple teams into a consensus prediction are illustrated by the rank sum prediction (triangles). Though no rank sum prediction scored higher than the best-performer, a consensus of the predictions of the second and third place teams boosted the score of the second place team. Rank sum analysis shown for the 100-node sub-challenge.
Figure 6 Analysis of the community of teams reveals characteristics of identifiable and unidentifiable network edges. The number of teams that identify an edge at a specified cutoff is a measure of how easy or difficult an edge is to identify. In this analysis we use a cutoff of 2P (i.e., twice the number of actual positive edges in the gold standard network). (a) Histograms indicate the number of teams that correctly identified the edges of the gold standard network called InSilico_Size100_Ecoli1. The ten worst teams in the 100-node sub-challenge identified about the same number of edges as is expected by chance. By contrast, the ten best teams identified more edges than is expected by chance and this sub-community has a markedly different identifiability distribution than random. Still, some edges were not identified by even the ten best teams (see bin corresponding to zero teams). Unidentified edges are characterized by (b) a property of the measurement data and (c) a topological property of the network. (b) Unidentified edges have a lower null-mutant absolute z-score than those that were identified by at least one of the ten best teams. This metric is a measure of the information content of the measurements. (c) Unidentified edges belong to target nodes with a higher in-degree than edges that were identified by at least one of the ten best teams. Circles denote the median and bars denote upper and lower quartiles. Statistics were not computed for bins containing less than four edges. (d) The benefits of combining the predictions of multiple teams into a consensus prediction are illustrated by the rank sum prediction (triangles). Though no rank sum prediction scored higher than the best-performer, a consensus of the predictions of the second and third place teams boosted the score of the second place team. Rank sum analysis shown for the 100-node sub-challenge.CC BY · PubMed Central
Figure 7 Community analysis of systematic false positives. Systematic false positive (FP) edges are the top one percent of edges that were predicted by the most teams to exist, yet are actually absent from the gold standard (i.e., negative). Rare false positive edges are the remaining 99 percent of edges that are absent from the gold standard network. The entries of each two-by-two contingency table sum to the total number of negative edges (i.e., those not present) in the gold standard network. There is a relative concentration of FP errors in the shortcut and co-regulated topologies, as evidenced by the A-to-B ratio. P-values for each contingency table were computed by Fisher's exact test, which expresses the probability that a random partitioning of the data will result in such a contingency table.
Figure 7 Community analysis of systematic false positives. Systematic false positive (FP) edges are the top one percent of edges that were predicted by the most teams to exist, yet are actually absent from the gold standard (i.e., negative). Rare false positive edges are the remaining 99 percent of edges that are absent from the gold standard network. The entries of each two-by-two contingency table sum to the total number of negative edges (i.e., those not present) in the gold standard network. There is a relative concentration of FP errors in the shortcut and co-regulated topologies, as evidenced by the A-to-B ratio. P-values for each contingency table were computed by Fisher's exact test, which expresses the probability that a random partitioning of the data will result in such a contingency table.CC BY · PubMed Central
Figure 8 Survey of in silico network methods. There does not seem to be a correlation between methods and scores, implying that success is more related to the details of implementation than the choice of general methodology.
Figure 8 Survey of in silico network methods. There does not seem to be a correlation between methods and scores, implying that success is more related to the details of implementation than the choice of general methodology.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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 33 references

Research works printed but not linked: 1

  1. [15]Alexopoulos L Saez-Rodriguez J Cosgrove B Lauffenburger D PK S 2009 Networks reconstructed from cell response data reveal profound dif ferences in signaling by tolllike receptors and nf-?b in normal and transformed human hepatocytes. Under review

Cited by

See all 439

Metadata

Authors
Prill, R. J., Marbach, D., Saez-Rodriguez, J., Sorger, P. K., Alexopoulos, L. G., Xue, X., Clarke, N. D., Altan-Bonnet, G., & Stolovitzky, G.
Affiliations
  • IBM (United States) (US)
  • École Polytechnique Fédérale de Lausanne (CH)
  • Harvard University (US)
  • National Technical University of Athens (GR)
  • Genome Institute of Singapore (SG)
Corresponding author
Stolovitzky, Gustavo
Venue
PLoS ONE
Published
2010-02-23
Volume
5
Issue
2
Pages
e9202
ISSN
1932-6203 (ISSN-L)
OpenAlex
W2143076232
Cited by
439
References
33
Citations by year
2026: 16 · 2025: 16 · 2024: 15 · 2023: 11 · 2022: 21 · 2021: 18 · 2020: 17 · 2019: 17 · 2018: 28 · 2017: 27 · 2016: 23 · 2015: 50 · 2014: 40 · 2013: 38 · 2012: 47
FWCI
18.23
Type
journal article
Language
en
Open access
Locations
  • PLoS ONE (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Digital Access to Scholarship at Harvard (DASH) (Harvard University) (submittedVersion, cc-by)
  • National University of Singapore (submittedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • DSpace - NTUA (National Technical University of Athens) (submittedVersion, other-oa)
  • Infoscience (Ecole Polytechnique Fédérale de Lausanne) (submittedVersion)
  • Infoscience (Ecole Polytechnique Fédérale de Lausanne) (submittedVersion, cc-by-nc-nd)
Funders
  • National Science Foundation (GM68762, 0848030)
  • National Institutes of Health (3p50gm068762-07s2, NIH P50, GM68762)
  • National Cancer Institute (U54CA112967)
  • National Institute of General Medical Sciences (P50 GM68762, P50-GM068762)
  • National Institute of Allergy and Infectious Diseases (R56AI083408, AI083408, R01AI083408)
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 › Microbial Metabolic Engineering and Bioproduction
Keywords
network inference, reverse engineering, gene expression prediction, ensemble forecasting, model validation
MeSH terms
Algorithms, Animals, Humans, Models, Biological, Software, Reproducibility of Results, Signal Transduction, Cluster Analysis, Computational Biology, Gene Expression Profiling, Protein Interaction Mapping, Systems Biology, Gene Regulatory Networks

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: Towards a Rigorous Assessment of Systems Biology Models: The DREAM3 Challenges. https://app.cochat.ai/papers/10.1371/journal.pone.0009202
The paper
Prill, R. J., Marbach, D., Saez-Rodriguez, J., Sorger, P. K., Alexopoulos, L. G., Xue, X., Clarke, N. D., Altan-Bonnet, G., & Stolovitzky, G. (2010). Towards a Rigorous Assessment of Systems Biology Models: The DREAM3 Challenges. PLoS ONE. https://doi.org/10.1371/journal.pone.0009202

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.