CoChat
Sign Up
Cited by 150Genomics and Rare DiseasesBioinformatics and Genomic NetworksBiomedical Text Mining and Ontologies

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

Improving the diagnostic yield of exome- sequencing by predicting gene–phenotype associations using large-scale gene expression analysis

Open access
Deelen, Patrick, van Dam, Sipko, Herkert, Johanna C., Karjalainen, Juha M., Brugge, Harm, Abbott, Kristin M., van Diemen, Cleo C., van der Zwaag, Paul A., Gerkes, Erica H., Zonneveld-Huijssoon, Evelien, Boer-Bergsma, Jelkje J., Folkertsma, Pytrik, Gillett, Tessa, van der Velde, K. Joeri, Kanninga, Roan, van den Akker, Peter C., Jan, Sabrina Z., Hoorntje, Edgar T., te Rijdt, Wouter P., Vos, Yvonne J., Jongbloed, Jan D. H., van Ravenswaaij-Arts, Conny M. A., Sinke, Richard, Sikkema-Raddatz, Birgit, Kerstjens-Frederikse, Wilhelmina S., Swertz, Morris A., Franke, Lude·2019·Nature Communications
DOI 10.1038/s41467-019-10649-4PMC PMC6599066

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

The diagnostic yield of exome and genome sequencing remains low (8–70%), due to incomplete knowledge on the genes that cause disease. To improve this, we use RNA-seq data from 31,499 samples to predict which genes cause specific disease phenotypes, and develop GeneNetwork Assisted Diagnostic Optimization (GADO). We show that this unbiased method, which does not rely upon specific knowledge on individual genes, is effective in both identifying previously unknown disease gene associations, and flagging genes that have previously been incorrectly implicated in disease. GADO can be run on www.genenetwork.nl by supplying HPO-terms and a list of genes that contain candidate variants. Finally, applying GADO to a cohort of 61 patients for whom exome-sequencing analysis had not resulted in a genetic diagnosis, yields likely causative genes for ten cases.

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/s41467-019-10649-4

Cites

57 of 57 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Schematic overview of GADO. a Per patient, GADO requires a set of phenotypic features (encoded using HPO terms) and a list of candidate genes (gene names either entered using HGNC symbols or Ensembl IDs). This gene list should contain genes in which rare variants have been observed for the patient. It then ascertains whether any of these genes have been predicted to cause the phenotypic features, observed in the patient. These HPO phenotypes predictions per gene are based on observed co-regulation with sets of genes that are already known to be associated with these phenotypes. b Overview of how disease symptoms are predicted using gene expression data from 31,499 human RNA-seq samples. A principal component analysis on the co-expression matrix results in the identification of 1588 significant principal components. For each HPO term we investigate every component: per component we test whether there is a significant difference between eigenvector coefficients of genes known to cause a specific phenotype and a background set of genes. This results in a matrix that indicates which principal components are informative for every HPO term. By correlating this matrix to the eigenvector coefficients of every individual gene, it is possible to infer the likely HPO disease phenotype term that would be the result of a pathogenic variant in that gene
Fig. 1 Schematic overview of GADO. a Per patient, GADO requires a set of phenotypic features (encoded using HPO terms) and a list of candidate genes (gene names either entered using HGNC symbols or Ensembl IDs). This gene list should contain genes in which rare variants have been observed for the patient. It then ascertains whether any of these genes have been predicted to cause the phenotypic features, observed in the patient. These HPO phenotypes predictions per gene are based on observed co-regulation with sets of genes that are already known to be associated with these phenotypes. b Overview of how disease symptoms are predicted using gene expression data from 31,499 human RNA-seq samples. A principal component analysis on the co-expression matrix results in the identification of 1588 significant principal components. For each HPO term we investigate every component: per component we test whether there is a significant difference between eigenvector coefficients of genes known to cause a specific phenotype and a background set of genes. This results in a matrix that indicates which principal components are informative for every HPO term. By correlating this matrix to the eigenvector coefficients of every individual gene, it is possible to infer the likely HPO disease phenotype term that would be the result of a pathogenic variant in that geneCC BY · PubMed Central
Fig. 2 A compendium of gene expression profiles that can be used for gene function prediction. We downloaded 31,499 RNA-seq samples from ENA. These samples come from many different studies. They show coherent clustering after correcting for technical biases. Generally, samples originating from the same tissue, cell-type or cell-line cluster together. The two axes denote the two t-SNE components. The number of samples per tissue or cell-type are mentioned, and after the colon the number of unique studies is mentioned, indicating that samples cluster by tissue or cell-type, and that this clustering is not due to systematic technical confounding due to the fact that for a given tissue, samples come from only a single laboratory
Fig. 2 A compendium of gene expression profiles that can be used for gene function prediction. We downloaded 31,499 RNA-seq samples from ENA. These samples come from many different studies. They show coherent clustering after correcting for technical biases. Generally, samples originating from the same tissue, cell-type or cell-line cluster together. The two axes denote the two t-SNE components. The number of samples per tissue or cell-type are mentioned, and after the colon the number of unique studies is mentioned, indicating that samples cluster by tissue or cell-type, and that this clustering is not due to systematic technical confounding due to the fact that for a given tissue, samples come from only a single laboratoryCC BY · PubMed Central
Fig. 3 Performance of disease gene prioritization compared to random permutation. a OMIM disease genes and provisional disease genes have significantly stronger prioritization Z-scores compared to permuted disease genes (T-test p-values: 2.16 × 10−532 and 5.38 × 10−80, respectively). We also observe that the predictions of the provisional OMIM genes are, on average, weaker than the other OMIM disease genes (T-test p-value: 1.89 × 10−7). Because we use a leave-one-out strategy when calculating prioritization Z-scores for genes that have already been associated to an HPO term, there is no prediction bias towards known associations. Therefore, this benchmark is informative of the power to predict novel associations (see methods). b We observe a significant relation (Spearman p-value: 1.01 × 10−4) between the burden of evidence that a gene is associated to a disease and the GADO prioritization Z-score. Most genes are scored by13 some additional refuted genes, denotated as squares or diamonds, are reported by ref. 8 and ref. 12 c We observe a clear relation between the prioritization Z-scores and the gene predictability scores (Pearson r = 0.54). We do not observe this relation in the permuted results. d Our gene prioritization Z-scores are significantly correlated (Pearson p-value: 1.67 × 10−23) to the number of likely pathogenic (LP) and pathogenic (P) variants reported for a gene in ClinVar
Fig. 3 Performance of disease gene prioritization compared to random permutation. a OMIM disease genes and provisional disease genes have significantly stronger prioritization Z-scores compared to permuted disease genes (T-test p-values: 2.16 × 10−532 and 5.38 × 10−80, respectively). We also observe that the predictions of the provisional OMIM genes are, on average, weaker than the other OMIM disease genes (T-test p-value: 1.89 × 10−7). Because we use a leave-one-out strategy when calculating prioritization Z-scores for genes that have already been associated to an HPO term, there is no prediction bias towards known associations. Therefore, this benchmark is informative of the power to predict novel associations (see methods). b We observe a significant relation (Spearman p-value: 1.01 × 10−4) between the burden of evidence that a gene is associated to a disease and the GADO prioritization Z-score. Most genes are scored by13 some additional refuted genes, denotated as squares or diamonds, are reported by ref. 8 and ref. 12 c We observe a clear relation between the prioritization Z-scores and the gene predictability scores (Pearson r = 0.54). We do not observe this relation in the permuted results. d Our gene prioritization Z-scores are significantly correlated (Pearson p-value: 1.67 × 10−23) to the number of likely pathogenic (LP) and pathogenic (P) variants reported for a gene in ClinVarCC BY · PubMed Central
Fig. 4 Performance of GeneNetwork on solved cases. a Comparison between using GADO and Exomiser to rank candidate variants. b Our cohort contained a case with two distinct conditions, and clustering showed the HPO terms of the same disease are closest to each other. Note, the HPO term “Inflammation of the large intestine” did not yield a significant prediction profile and therefore the parent terms “Abnormality of the large intestine”, “Increased inflammatory response” and “Functional abnormality of the gastrointestinal tract” were used for this case
Fig. 4 Performance of GeneNetwork on solved cases. a Comparison between using GADO and Exomiser to rank candidate variants. b Our cohort contained a case with two distinct conditions, and clustering showed the HPO terms of the same disease are closest to each other. Note, the HPO term “Inflammation of the large intestine” did not yield a significant prediction profile and therefore the parent terms “Abnormality of the large intestine”, “Increased inflammatory response” and “Functional abnormality of the gastrointestinal tract” were used for this caseCC BY · PubMed Central
Fig. 5 a Prioritization results of one of our previously solved cases (www.genenetwork.nl). This patient was diagnosed with Kleefstra syndrome. The patient only showed a few of the phenotypic features associated with Kleefstra syndrome and additionally had a neoplasm of the pituitary (which is not associated with Kleefstra syndrome). Despite this limited overlap in phenotypic features, GADO was able to rank the causative gene (EHMT1) second. Here, we also show the value of the HPO clustering heatmap: the two terms related to the neoplasm cluster separately from the intellectual disability and the facial abnormalities that are associated to Kleefstra syndrome. b Clustering of a set of genes allowing function/HPO enrichment of all genes or specific enrichment of automatically defined sub clusters. Here, we loaded all known DCM genes and OBSCN, and we focus on a sub-cluster of genes containing OBSCN (highlighted by the arrow). We see that it is strongly co-regulated with many of the known DCM genes. Pathway enrichment of this sub-cluster reveals that these genes are most strongly enriched for the muscle contraction Reactome pathway. DCM, Dilated Cardiomyopathy
Fig. 5 a Prioritization results of one of our previously solved cases (www.genenetwork.nl). This patient was diagnosed with Kleefstra syndrome. The patient only showed a few of the phenotypic features associated with Kleefstra syndrome and additionally had a neoplasm of the pituitary (which is not associated with Kleefstra syndrome). Despite this limited overlap in phenotypic features, GADO was able to rank the causative gene (EHMT1) second. Here, we also show the value of the HPO clustering heatmap: the two terms related to the neoplasm cluster separately from the intellectual disability and the facial abnormalities that are associated to Kleefstra syndrome. b Clustering of a set of genes allowing function/HPO enrichment of all genes or specific enrichment of automatically defined sub clusters. Here, we loaded all known DCM genes and OBSCN, and we focus on a sub-cluster of genes containing OBSCN (highlighted by the arrow). We see that it is strongly co-regulated with many of the known DCM genes. Pathway enrichment of this sub-cluster reveals that these genes are most strongly enriched for the muscle contraction Reactome pathway. DCM, Dilated CardiomyopathyCC 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

Research works printed but not linked: 1

  1. Fehrmann, R. S. N. et al. Gene expression analysis identifies global gene dosage sensitivity in cancer. Nat. Genet. 47, 115–125 (2015). · DOI · Paper page

Cited by

Sorted by citations

See all 150

Related

Metadata

Authors
Deelen, P., van Dam, S., Herkert, J. C., Karjalainen, J. M., Brugge, H., Abbott, K. M., van Diemen, C. C., van der Zwaag, P. A., Gerkes, E. H., Zonneveld-Huijssoon, E., Boer-Bergsma, J. J., Folkertsma, P., Gillett, T., van der Velde, K. J., Kanninga, R., van den Akker, P. C., Jan, S. Z., Hoorntje, E. T., te Rijdt, W. P., Vos, Y. J., Jongbloed, J. D. H., van Ravenswaaij-Arts, C. M. A., Sinke, R., Sikkema-Raddatz, B., Kerstjens-Frederikse, W. S., Swertz, M. A., & Franke, L.
Affiliations
University Medical Center Groningen (NL)
Corresponding author
Franke, Lude
ORCID
Venue
Nature Communications
Publisher
Springer Science and Business Media LLC
Published
2019-06-28
Volume
10
Issue
1
Pages
2837-2837
ISSN
2041-1723 (ISSN-L)
DOI
10.1038/s41467-019-10649-4
PMC
PMC6599066
OpenAlex
W2953983664
Cited by
138
References
57
Citations by year
2026: 6 · 2025: 7 · 2024: 15 · 2023: 17 · 2022: 26 · 2021: 38 · 2020: 18 · 2019: 9
FWCI
13.67
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • University of Groningen research database (University of Groningen / Centre for Information Technology) (acceptedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
  • University of Groningen research database (University of Groningen / Centre for Information Technology) (submittedVersion, cc-by)
  • University of Groningen research database (University of Groningen / Centre for Information Technology) (submittedVersion)
Funders
University of Groningen
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Genomics and Rare Diseases
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
  • Health Sciences › Medicine › Health Informatics › Biomedical Text Mining and Ontologies
Keywords
exome sequencing, diagnostic yield, genotype-phenotype correlation, RNA-seq, gene expression, disease gene discovery, Human Phenotype Ontology, candidate variants, genetic diagnosis
MeSH terms
Gene Expression Regulation, Humans, Models, Genetic, Software, User-Computer Interface, Sequence Analysis, RNA, Genetic Predisposition to Disease, Principal Component Analysis, Databases, Nucleic Acid, 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