CoChat

On the inconsistency of ℓ 1-penalised sparse precision matrix estimation

Open access

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

Abstract

Various ℓ 1-penalised estimation methods such as graphical lasso and CLIME are widely used for sparse precision matrix estimation and learning of undirected network structure from data. Many of these methods have been shown to be consistent under various quantitative assumptions about the underlying true covariance matrix. Intuitively, these conditions are related to situations where the penalty term will dominate the optimisation. We explore the consistency of ℓ 1-based methods for a class of bipartite graphs motivated by the structure of models commonly used for gene regulatory networks. We show that all ℓ 1-based methods fail dramatically for models with nearly linear dependencies between the variables. We also study the consistency on models derived from real gene expression data and note that the assumptions needed for consistency never hold even for modest sized gene networks and ℓ 1-based methods also become unreliable in practice for larger networks. Our results demonstrate that ℓ 1-penalised undirected network structure learning methods are unable to reliably learn many sparse bipartite graph structures, which arise often in gene expression data. Users of such methods should be aware of the consistency criteria of the methods and check if they are likely to be met in their application of interest.

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 · 10.1186/s12859-016-1309-x

Cites

33 of 33 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Left: Graphical representation of a latent variable model as an undirected graphical model for a case with somewhat sparse A. Right: The adjacency matrix of the graph showing the sparse pattern of non-zero elements in the corresponding precision matrix
Fig. 1 Left: Graphical representation of a latent variable model as an undirected graphical model for a case with somewhat sparse A. Right: The adjacency matrix of the graph showing the sparse pattern of non-zero elements in the corresponding precision matrixCC BY · PubMed Central
Fig. 2 Performances of different methods on the bipartite graph model with 100 samples. (Lower values are better)
Fig. 2 Performances of different methods on the bipartite graph model with 100 samples. (Lower values are better)CC BY · PubMed Central
Fig. 3 Performances of different methods on the bipartite graph model with 1000 samples. (Lower values are better)
Fig. 3 Performances of different methods on the bipartite graph model with 1000 samples. (Lower values are better)CC BY · PubMed Central
Fig. 4 Performances of different methods on the bipartite graph model with varying output dimensionality. (Lower values are better)
Fig. 4 Performances of different methods on the bipartite graph model with varying output dimensionality. (Lower values are better)CC BY · PubMed Central
Fig. 5 Performances of different methods on the bipartite graph model with varying input dimensionality. (Lower values are better)
Fig. 5 Performances of different methods on the bipartite graph model with varying input dimensionality. (Lower values are better)CC BY · PubMed Central
Fig. 6 Contributions of the different terms of the glasso objective (1) for the bipartite graph model with 1000 samples. The green curves show the contributions of the first two terms of Eq. (1) and the blue curves show the contributions of the last penalty term. Solid lines show the result of the glasso optimal solution while dashed lines show the result for the true solution
Fig. 6 Contributions of the different terms of the glasso objective (1) for the bipartite graph model with 1000 samples. The green curves show the contributions of the first two terms of Eq. (1) and the blue curves show the contributions of the last penalty term. Solid lines show the result of the glasso optimal solution while dashed lines show the result for the true solutionCC BY · PubMed Central
Fig. 7 Precision of glasso on infinite data as a function of the norm γ of Assumption 1 and Eq. (3). Values to the left of the green vertical line satisfy this condition while values to the right violate it. (Higher values are better)
Fig. 7 Precision of glasso on infinite data as a function of the norm γ of Assumption 1 and Eq. (3). Values to the left of the green vertical line satisfy this condition while values to the right violate it. (Higher values are better)CC BY · PubMed Central
Fig. 8 Testing the condition of Assumption 1 of [22] in Eq. (3) on real gene expression data showing the fraction of random subsets of d genes that fulfil the requirement and various relaxations. The condition (3) requires γ<1, but the figure shows results also for larger γ cutoffs, denoted by c
Fig. 8 Testing the condition of Assumption 1 of [22] in Eq. (3) on real gene expression data showing the fraction of random subsets of d genes that fulfil the requirement and various relaxations. The condition (3) requires γ<1, but the figure shows results also for larger γ cutoffs, denoted by cCC BY · PubMed Central
Fig. 9 Average precisions for glasso with different dimensions and sample sizes of the real gene expression data, higher values are better. In these experiments, 50 data sets were created. We encountered convergence problems with few of the data sets and the corresponding results were omitted when computing the average values shown here. The precision obtained by random guessing is also illustrated
Fig. 9 Average precisions for glasso with different dimensions and sample sizes of the real gene expression data, higher values are better. In these experiments, 50 data sets were created. We encountered convergence problems with few of the data sets and the corresponding results were omitted when computing the average values shown here. The precision obtained by random guessing is also illustratedCC BY · PubMed Central
Fig. 10 Average contributions of the different terms of the glasso objective function (1) on real gene expression data over random subsets of d genes. The values are shown for the ℓ 1 penalty term as well as the unnormalised log-likelihood, divided by d to make them comparable. Solid lines show the values for glasso result while dashed lines show the result for ground truth
Fig. 10 Average contributions of the different terms of the glasso objective function (1) on real gene expression data over random subsets of d genes. The values are shown for the ℓ 1 penalty term as well as the unnormalised log-likelihood, divided by d to make them comparable. Solid lines show the values for glasso result while dashed lines show the result for ground truthCC 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

Cited by

See all 14

Metadata

Authors
Heinävaara, O., Leppä-aho, J., Corander, J., & Honkela, A.
Affiliations
University of Helsinki (FI)
Corresponding author
Honkela, Antti
Venue
BMC Bioinformatics
Published
2016-12
Volume
17
Issue
S16
Pages
448-448
ISSN
1471-2105 (ISSN-L)
OpenAlex
W2296395953
Cited by
14
References
33
Citations by year
2026: 1 · 2025: 3 · 2024: 1 · 2023: 1 · 2021: 1 · 2020: 1 · 2019: 2 · 2018: 1 · 2017: 2 · 2016: 1
FWCI
2.02
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • BMC Bioinformatics (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Työväentutkimus Vuosikirja (submittedVersion)
Funders
  • Academy of Finland (259440, 251170)
  • FP7 Ideas: European Research Council (239784)
Topics
  • Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Inference
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene expression and cancer classification
Keywords
sparse precision matrix estimation, consistency, graphical lasso, bipartite graphs, gene regulatory networks, gene expression data, linear dependencies, covariance matrix estimation
MeSH terms
Machine Learning, Animals, Humans, Models, Statistical, Computational Biology, Gene Regulatory Networks, Transcriptome

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: On the inconsistency of ℓ 1-penalised sparse precision matrix estimation. https://app.cochat.ai/papers/10.1186/s12859-016-1309-x
The paper
Heinävaara, O., Leppä-aho, J., Corander, J., & Honkela, A. (2016). On the inconsistency of ℓ 1-penalised sparse precision matrix estimation. BMC Bioinformatics. https://doi.org/10.1186/s12859-016-1309-x

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.