Limits...
Assessing the biological significance of gene expression signatures and co-expression modules by studying their network properties.

Minguez P, Dopazo J - PLoS ONE (2011)

Bottom Line: Paradoxically, in spite of the apparent functional role that might be attributed to such gene sets, signatures do not seem to be reproducible across experiments.A total of 3904 gene ontology (GO) modules, 146 KEGG pathways, and 263 Biocarta pathways have been used as functional modules of reference.Co-expression modules found in microarray experiments display a high level of connectivity, similar to the one shown by conventional modules based on functional definitions (GO, KEGG and Biocarta).These results suggest that gene selection methods which take into account relationships among genes should be superior to methods that assume independence among genes outside their functional contexts.

View Article: PubMed Central - PubMed

Affiliation: Department of Bioinformatics and Genomics, Centro de Investigación Príncipe Felipe, Valencia, Spain.

ABSTRACT
Microarray experiments have been extensively used to define signatures, which are sets of genes that can be considered markers of experimental conditions (typically diseases). Paradoxically, in spite of the apparent functional role that might be attributed to such gene sets, signatures do not seem to be reproducible across experiments. Given the close relationship between function and protein interaction, network properties can be used to study to what extent signatures are composed of genes whose resulting proteins show a considerable level of interaction (and consequently a putative common functional role).We have analysed 618 signatures and 507 modules of co-expression in cancer looking for significant values of four main protein-protein interaction (PPI) network parameters: connection degree, cluster coefficient, betweenness and number of components. A total of 3904 gene ontology (GO) modules, 146 KEGG pathways, and 263 Biocarta pathways have been used as functional modules of reference.Co-expression modules found in microarray experiments display a high level of connectivity, similar to the one shown by conventional modules based on functional definitions (GO, KEGG and Biocarta). A general observation for all the classes studied is that the networks formed by the modules improve their topological parameters when an external protein is allowed to be introduced within the paths (up to the 70% of GO modules show network parameters beyond the random expectation). This fact suggests that functional definitions are incomplete and some genes might still be missing. Conversely, signatures are clearly not capturing the altered functions in the corresponding studies. This is probably because the way in which the genes have been selected in the signatures is too conservative. These results suggest that gene selection methods which take into account relationships among genes should be superior to methods that assume independence among genes outside their functional contexts.

Show MeSH

Related in: MedlinePlus

Local properties of the nodes represented by the different network parameters used.Connection degree was computed as the number of edges (interaction events) for a given node. A shows a node with a high connectivity. B has both high clustering coefficient and connectivity. C has a high betweenness. It is a hub because many shortest paths joining nodes pass through it. The connecting edge between two component (D) is known as the articulation point.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3049771&req=5

pone-0017474-g005: Local properties of the nodes represented by the different network parameters used.Connection degree was computed as the number of edges (interaction events) for a given node. A shows a node with a high connectivity. B has both high clustering coefficient and connectivity. C has a high betweenness. It is a hub because many shortest paths joining nodes pass through it. The connecting edge between two component (D) is known as the articulation point.

Mentions: Different network parameters represent local and global network properties. Figure 5 schematizes the local properties of the nodes portrayed by the different parameters used. The properties used in this study are: Connection degree, Clustering coefficient, Betweenness centrality and components.


Assessing the biological significance of gene expression signatures and co-expression modules by studying their network properties.

Minguez P, Dopazo J - PLoS ONE (2011)

Local properties of the nodes represented by the different network parameters used.Connection degree was computed as the number of edges (interaction events) for a given node. A shows a node with a high connectivity. B has both high clustering coefficient and connectivity. C has a high betweenness. It is a hub because many shortest paths joining nodes pass through it. The connecting edge between two component (D) is known as the articulation point.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3049771&req=5

pone-0017474-g005: Local properties of the nodes represented by the different network parameters used.Connection degree was computed as the number of edges (interaction events) for a given node. A shows a node with a high connectivity. B has both high clustering coefficient and connectivity. C has a high betweenness. It is a hub because many shortest paths joining nodes pass through it. The connecting edge between two component (D) is known as the articulation point.
Mentions: Different network parameters represent local and global network properties. Figure 5 schematizes the local properties of the nodes portrayed by the different parameters used. The properties used in this study are: Connection degree, Clustering coefficient, Betweenness centrality and components.

Bottom Line: Paradoxically, in spite of the apparent functional role that might be attributed to such gene sets, signatures do not seem to be reproducible across experiments.A total of 3904 gene ontology (GO) modules, 146 KEGG pathways, and 263 Biocarta pathways have been used as functional modules of reference.Co-expression modules found in microarray experiments display a high level of connectivity, similar to the one shown by conventional modules based on functional definitions (GO, KEGG and Biocarta).These results suggest that gene selection methods which take into account relationships among genes should be superior to methods that assume independence among genes outside their functional contexts.

View Article: PubMed Central - PubMed

Affiliation: Department of Bioinformatics and Genomics, Centro de Investigación Príncipe Felipe, Valencia, Spain.

ABSTRACT
Microarray experiments have been extensively used to define signatures, which are sets of genes that can be considered markers of experimental conditions (typically diseases). Paradoxically, in spite of the apparent functional role that might be attributed to such gene sets, signatures do not seem to be reproducible across experiments. Given the close relationship between function and protein interaction, network properties can be used to study to what extent signatures are composed of genes whose resulting proteins show a considerable level of interaction (and consequently a putative common functional role).We have analysed 618 signatures and 507 modules of co-expression in cancer looking for significant values of four main protein-protein interaction (PPI) network parameters: connection degree, cluster coefficient, betweenness and number of components. A total of 3904 gene ontology (GO) modules, 146 KEGG pathways, and 263 Biocarta pathways have been used as functional modules of reference.Co-expression modules found in microarray experiments display a high level of connectivity, similar to the one shown by conventional modules based on functional definitions (GO, KEGG and Biocarta). A general observation for all the classes studied is that the networks formed by the modules improve their topological parameters when an external protein is allowed to be introduced within the paths (up to the 70% of GO modules show network parameters beyond the random expectation). This fact suggests that functional definitions are incomplete and some genes might still be missing. Conversely, signatures are clearly not capturing the altered functions in the corresponding studies. This is probably because the way in which the genes have been selected in the signatures is too conservative. These results suggest that gene selection methods which take into account relationships among genes should be superior to methods that assume independence among genes outside their functional contexts.

Show MeSH
Related in: MedlinePlus