Limits...
Robust detection of hierarchical communities from Escherichia coli gene expression data.

Treviño S, Sun Y, Cooper TF, Bassler KE - PLoS Comput. Biol. (2012)

Bottom Line: However, commonly used algorithms can yield unreliable results due to experimental noise, algorithmic stochasticity, and the influence of arbitrarily chosen parameter values.These communities significantly enrich for gene ontology (GO) terms, consistent with them representing biologically meaningful groups.Further, analysis of the most significantly enriched communities identified several candidate new regulatory interactions.

View Article: PubMed Central - PubMed

Affiliation: Department of Physics, University of Houston, Houston, Texas, United States of America.

ABSTRACT
Determining the functional structure of biological networks is a central goal of systems biology. One approach is to analyze gene expression data to infer a network of gene interactions on the basis of their correlated responses to environmental and genetic perturbations. The inferred network can then be analyzed to identify functional communities. However, commonly used algorithms can yield unreliable results due to experimental noise, algorithmic stochasticity, and the influence of arbitrarily chosen parameter values. Furthermore, the results obtained typically provide only a simplistic view of the network partitioned into disjoint communities and provide no information of the relationship between communities. Here, we present methods to robustly detect co-regulated and functionally enriched gene communities and demonstrate their application and validity for Escherichia coli gene expression data. Applying a recently developed community detection algorithm to the network of interactions identified with the context likelihood of relatedness (CLR) method, we show that a hierarchy of network communities can be identified. These communities significantly enrich for gene ontology (GO) terms, consistent with them representing biologically meaningful groups. Further, analysis of the most significantly enriched communities identified several candidate new regulatory interactions. The robustness of our methods is demonstrated by showing that a core set of functional communities is reliably found when artificial noise, modeling experimental noise, is added to the data. We find that noise mainly acts conservatively, increasing the relatedness required for a network link to be reliably assigned and decreasing the size of the core communities, rather than causing association of genes into new communities.

Show MeSH

Related in: MedlinePlus

The effect of noise on core community structure and GO term enrichment.(A) Proportion of  core community nodes that remain in a core community. (B) The number of significant GO term enrichments as a function of noise level  for networks constructed with . If a GO term is enriched by more than one community, each enrichment is counted separately.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3285575&req=5

pcbi-1002391-g004: The effect of noise on core community structure and GO term enrichment.(A) Proportion of core community nodes that remain in a core community. (B) The number of significant GO term enrichments as a function of noise level for networks constructed with . If a GO term is enriched by more than one community, each enrichment is counted separately.

Mentions: For each noisy data set, we used the CLR algorithm to infer a regulation network at an value of 2, and the community structure was determined with the methods described above. For each dataset, 10 different community partitionings were obtained, giving a total of 200 partititonings for each value of . Figure 3 shows a series of correlation matrix plots for the community structure found for the partitioning ensembles for . The degree of noise clearly has a major impact on community structure. Nevertheless, except at , there exist robustly determined core communities. In addition this analysis revealed two important results. First, as the noise level increased, a large proportion of the genes in a core community are partitioned into sub communities but genes rarely switch out of their core communities. This is similar to what happens when the threshold value for creating the network was increased (Figure 2). Second, with one exception, the number of nodes included in each core community decreased as was increased (Figure 4A). We conclude that noise acts mainly conservatively, decreasing the size of core communities, rather than causing association of genes into new communities.


Robust detection of hierarchical communities from Escherichia coli gene expression data.

Treviño S, Sun Y, Cooper TF, Bassler KE - PLoS Comput. Biol. (2012)

The effect of noise on core community structure and GO term enrichment.(A) Proportion of  core community nodes that remain in a core community. (B) The number of significant GO term enrichments as a function of noise level  for networks constructed with . If a GO term is enriched by more than one community, each enrichment is counted separately.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3285575&req=5

pcbi-1002391-g004: The effect of noise on core community structure and GO term enrichment.(A) Proportion of core community nodes that remain in a core community. (B) The number of significant GO term enrichments as a function of noise level for networks constructed with . If a GO term is enriched by more than one community, each enrichment is counted separately.
Mentions: For each noisy data set, we used the CLR algorithm to infer a regulation network at an value of 2, and the community structure was determined with the methods described above. For each dataset, 10 different community partitionings were obtained, giving a total of 200 partititonings for each value of . Figure 3 shows a series of correlation matrix plots for the community structure found for the partitioning ensembles for . The degree of noise clearly has a major impact on community structure. Nevertheless, except at , there exist robustly determined core communities. In addition this analysis revealed two important results. First, as the noise level increased, a large proportion of the genes in a core community are partitioned into sub communities but genes rarely switch out of their core communities. This is similar to what happens when the threshold value for creating the network was increased (Figure 2). Second, with one exception, the number of nodes included in each core community decreased as was increased (Figure 4A). We conclude that noise acts mainly conservatively, decreasing the size of core communities, rather than causing association of genes into new communities.

Bottom Line: However, commonly used algorithms can yield unreliable results due to experimental noise, algorithmic stochasticity, and the influence of arbitrarily chosen parameter values.These communities significantly enrich for gene ontology (GO) terms, consistent with them representing biologically meaningful groups.Further, analysis of the most significantly enriched communities identified several candidate new regulatory interactions.

View Article: PubMed Central - PubMed

Affiliation: Department of Physics, University of Houston, Houston, Texas, United States of America.

ABSTRACT
Determining the functional structure of biological networks is a central goal of systems biology. One approach is to analyze gene expression data to infer a network of gene interactions on the basis of their correlated responses to environmental and genetic perturbations. The inferred network can then be analyzed to identify functional communities. However, commonly used algorithms can yield unreliable results due to experimental noise, algorithmic stochasticity, and the influence of arbitrarily chosen parameter values. Furthermore, the results obtained typically provide only a simplistic view of the network partitioned into disjoint communities and provide no information of the relationship between communities. Here, we present methods to robustly detect co-regulated and functionally enriched gene communities and demonstrate their application and validity for Escherichia coli gene expression data. Applying a recently developed community detection algorithm to the network of interactions identified with the context likelihood of relatedness (CLR) method, we show that a hierarchy of network communities can be identified. These communities significantly enrich for gene ontology (GO) terms, consistent with them representing biologically meaningful groups. Further, analysis of the most significantly enriched communities identified several candidate new regulatory interactions. The robustness of our methods is demonstrated by showing that a core set of functional communities is reliably found when artificial noise, modeling experimental noise, is added to the data. We find that noise mainly acts conservatively, increasing the relatedness required for a network link to be reliably assigned and decreasing the size of the core communities, rather than causing association of genes into new communities.

Show MeSH
Related in: MedlinePlus