Limits...
Modular biological function is most effectively captured by combining molecular interaction data types.

Ames RM, Macpherson JI, Pinney JW, Lovell SC, Robertson DL - PLoS ONE (2013)

Bottom Line: Furthermore, the different annotation types of GO are not predominantly associated with one of the interaction data types.Collectively our results demonstrate that successful capture of functional relationships by network data depends on both the specific biological function being characterised and the type of network data being used.Combining interaction subnetworks across data types is therefore essential for fully understanding the complex and emergent nature of biological function.

View Article: PubMed Central - PubMed

Affiliation: Computational and Evolutionary Biology, Faculty of Life Sciences, The University of Manchester, Manchester, United Kingdom. ryan.ames@manchester.ac.uk

ABSTRACT
Large-scale molecular interaction data sets have the potential to provide a comprehensive, system-wide understanding of biological function. Although individual molecules can be promiscuous in terms of their contribution to function, molecular functions emerge from the specific interactions of molecules giving rise to modular organisation. As functions often derive from a range of mechanisms, we demonstrate that they are best studied using networks derived from different sources. Implementing a graph partitioning algorithm we identify subnetworks in yeast protein-protein interaction (PPI), genetic interaction and gene co-regulation networks. Among these subnetworks we identify cohesive subgraphs that we expect to represent functional modules in the different data types. We demonstrate significant overlap between the subgraphs generated from the different data types and show these overlaps can represent related functions as represented by the Gene Ontology (GO). Next, we investigate the correspondence between our subgraphs and the Gene Ontology. This revealed varying degrees of coverage of the biological process, molecular function and cellular component ontologies, dependent on the data type. For example, subgraphs from the PPI show enrichment for 84%, 58% and 93% of annotated GO terms, respectively. Integrating the interaction data into a combined network increases the coverage of GO. Furthermore, the different annotation types of GO are not predominantly associated with one of the interaction data types. Collectively our results demonstrate that successful capture of functional relationships by network data depends on both the specific biological function being characterised and the type of network data being used. We identify functions that require integrated information to be accurately represented, demonstrating the limitations of individual data types. Combining interaction subnetworks across data types is therefore essential for fully understanding the complex and emergent nature of biological function.

Show MeSH

Related in: MedlinePlus

Composite functional maps.(A) GLASS visualisation in which each cell represents a GO term, coloured according to subgraphs that have the highest MCC for the enriched term. Blue, red, yellow and green colours indicate the subgraph with the highest MCC is from the PPI, genetic, co-regulation or combined network, respectively. Grey coloured cells are GO terms which have only one or no associated genes in any network. Areas ringed in black show examples of areas of the ontology which are best characterised by a single network. Panels B–C show examples of the best characterised subgraphs between all networks: (B) The mitochondrial small ribosomal subunit GO term is best represented by a subgraph from the PPI network. (C) A genetic subgraph best represents the IMP biosynthetic process GO term. (D) The GO term, SNAP receptor activity, is best represented by a subgraph in the combined network, created from all nodes and edges in the PPI, genetic and co-regulation networks. Nodes are coloured blue, red or green if they are present in the PPI, genetic or combined network, respectively, and are associated with each enriched GO term. White nodes represent nodes in a subgraph that are not associated with the enriched GO term. Edges are coloured blue, red, yellow or green if they are present in the PPI, genetic, co-regulation or combined network, respectively.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3643936&req=5

pone-0062670-g005: Composite functional maps.(A) GLASS visualisation in which each cell represents a GO term, coloured according to subgraphs that have the highest MCC for the enriched term. Blue, red, yellow and green colours indicate the subgraph with the highest MCC is from the PPI, genetic, co-regulation or combined network, respectively. Grey coloured cells are GO terms which have only one or no associated genes in any network. Areas ringed in black show examples of areas of the ontology which are best characterised by a single network. Panels B–C show examples of the best characterised subgraphs between all networks: (B) The mitochondrial small ribosomal subunit GO term is best represented by a subgraph from the PPI network. (C) A genetic subgraph best represents the IMP biosynthetic process GO term. (D) The GO term, SNAP receptor activity, is best represented by a subgraph in the combined network, created from all nodes and edges in the PPI, genetic and co-regulation networks. Nodes are coloured blue, red or green if they are present in the PPI, genetic or combined network, respectively, and are associated with each enriched GO term. White nodes represent nodes in a subgraph that are not associated with the enriched GO term. Edges are coloured blue, red, yellow or green if they are present in the PPI, genetic, co-regulation or combined network, respectively.

Mentions: Creation of a composite tree-map (Figure 5A), where the cell colours represent the network from which the terms are most accurately captured, allows direct comparison. From the trees on which the maps are based, we can identify distinct areas within an ontology, that are best characterised by subgraphs from a single network. Examples of such areas are outlined in Figure 5A. We can identify specific subgraphs from a single network that accurately characterise a single GO term. In the PPI network a single subgraph represents the mitochondrial small ribosomal subunit cellular component term, where 28/30 members annotated with the term and a MCC of 0.94 (Figure 5B). From the genetic network we have identified a subgraph that accurately represents the Inosine monophosphate (IMP) biosynthetic pathway and enzymes representative of the purine biosynthesis pathway (Figure 5C). These findings demonstrate that different areas of biological function are best represented by different types of biological data.


Modular biological function is most effectively captured by combining molecular interaction data types.

Ames RM, Macpherson JI, Pinney JW, Lovell SC, Robertson DL - PLoS ONE (2013)

Composite functional maps.(A) GLASS visualisation in which each cell represents a GO term, coloured according to subgraphs that have the highest MCC for the enriched term. Blue, red, yellow and green colours indicate the subgraph with the highest MCC is from the PPI, genetic, co-regulation or combined network, respectively. Grey coloured cells are GO terms which have only one or no associated genes in any network. Areas ringed in black show examples of areas of the ontology which are best characterised by a single network. Panels B–C show examples of the best characterised subgraphs between all networks: (B) The mitochondrial small ribosomal subunit GO term is best represented by a subgraph from the PPI network. (C) A genetic subgraph best represents the IMP biosynthetic process GO term. (D) The GO term, SNAP receptor activity, is best represented by a subgraph in the combined network, created from all nodes and edges in the PPI, genetic and co-regulation networks. Nodes are coloured blue, red or green if they are present in the PPI, genetic or combined network, respectively, and are associated with each enriched GO term. White nodes represent nodes in a subgraph that are not associated with the enriched GO term. Edges are coloured blue, red, yellow or green if they are present in the PPI, genetic, co-regulation or combined network, respectively.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3643936&req=5

pone-0062670-g005: Composite functional maps.(A) GLASS visualisation in which each cell represents a GO term, coloured according to subgraphs that have the highest MCC for the enriched term. Blue, red, yellow and green colours indicate the subgraph with the highest MCC is from the PPI, genetic, co-regulation or combined network, respectively. Grey coloured cells are GO terms which have only one or no associated genes in any network. Areas ringed in black show examples of areas of the ontology which are best characterised by a single network. Panels B–C show examples of the best characterised subgraphs between all networks: (B) The mitochondrial small ribosomal subunit GO term is best represented by a subgraph from the PPI network. (C) A genetic subgraph best represents the IMP biosynthetic process GO term. (D) The GO term, SNAP receptor activity, is best represented by a subgraph in the combined network, created from all nodes and edges in the PPI, genetic and co-regulation networks. Nodes are coloured blue, red or green if they are present in the PPI, genetic or combined network, respectively, and are associated with each enriched GO term. White nodes represent nodes in a subgraph that are not associated with the enriched GO term. Edges are coloured blue, red, yellow or green if they are present in the PPI, genetic, co-regulation or combined network, respectively.
Mentions: Creation of a composite tree-map (Figure 5A), where the cell colours represent the network from which the terms are most accurately captured, allows direct comparison. From the trees on which the maps are based, we can identify distinct areas within an ontology, that are best characterised by subgraphs from a single network. Examples of such areas are outlined in Figure 5A. We can identify specific subgraphs from a single network that accurately characterise a single GO term. In the PPI network a single subgraph represents the mitochondrial small ribosomal subunit cellular component term, where 28/30 members annotated with the term and a MCC of 0.94 (Figure 5B). From the genetic network we have identified a subgraph that accurately represents the Inosine monophosphate (IMP) biosynthetic pathway and enzymes representative of the purine biosynthesis pathway (Figure 5C). These findings demonstrate that different areas of biological function are best represented by different types of biological data.

Bottom Line: Furthermore, the different annotation types of GO are not predominantly associated with one of the interaction data types.Collectively our results demonstrate that successful capture of functional relationships by network data depends on both the specific biological function being characterised and the type of network data being used.Combining interaction subnetworks across data types is therefore essential for fully understanding the complex and emergent nature of biological function.

View Article: PubMed Central - PubMed

Affiliation: Computational and Evolutionary Biology, Faculty of Life Sciences, The University of Manchester, Manchester, United Kingdom. ryan.ames@manchester.ac.uk

ABSTRACT
Large-scale molecular interaction data sets have the potential to provide a comprehensive, system-wide understanding of biological function. Although individual molecules can be promiscuous in terms of their contribution to function, molecular functions emerge from the specific interactions of molecules giving rise to modular organisation. As functions often derive from a range of mechanisms, we demonstrate that they are best studied using networks derived from different sources. Implementing a graph partitioning algorithm we identify subnetworks in yeast protein-protein interaction (PPI), genetic interaction and gene co-regulation networks. Among these subnetworks we identify cohesive subgraphs that we expect to represent functional modules in the different data types. We demonstrate significant overlap between the subgraphs generated from the different data types and show these overlaps can represent related functions as represented by the Gene Ontology (GO). Next, we investigate the correspondence between our subgraphs and the Gene Ontology. This revealed varying degrees of coverage of the biological process, molecular function and cellular component ontologies, dependent on the data type. For example, subgraphs from the PPI show enrichment for 84%, 58% and 93% of annotated GO terms, respectively. Integrating the interaction data into a combined network increases the coverage of GO. Furthermore, the different annotation types of GO are not predominantly associated with one of the interaction data types. Collectively our results demonstrate that successful capture of functional relationships by network data depends on both the specific biological function being characterised and the type of network data being used. We identify functions that require integrated information to be accurately represented, demonstrating the limitations of individual data types. Combining interaction subnetworks across data types is therefore essential for fully understanding the complex and emergent nature of biological function.

Show MeSH
Related in: MedlinePlus