Limits...
Intertwining threshold settings, biological data and database knowledge to optimize the selection of differentially expressed genes from microarray.

Chuchana P, Holzmuller P, Vezilier F, Berthier D, Chantal I, Severac D, Lemesre JL, Cuny G, Nirdé P, Bucheton B - PLoS ONE (2010)

Bottom Line: Analysis performed during iterations helped us to select the optimal threshold required for the most pertinent selection of differentially expressed genes.We have applied this approach to the well documented mechanism of human macrophage response to lipopolysaccharide stimulation.We thus verified that our method was able to determine with the highest degree of accuracy the best threshold for selecting genes that are truly differentially expressed.

View Article: PubMed Central - PubMed

Affiliation: INSERM, Unité 844 - Montpellier, France. paul.chuchana@inserm.fr

ABSTRACT

Background: Many tools used to analyze microarrays in different conditions have been described. However, the integration of deregulated genes within coherent metabolic pathways is lacking. Currently no objective selection criterion based on biological functions exists to determine a threshold demonstrating that a gene is indeed differentially expressed.

Methodology/principal findings: To improve transcriptomic analysis of microarrays, we propose a new statistical approach that takes into account biological parameters. We present an iterative method to optimise the selection of differentially expressed genes in two experimental conditions. The stringency level of gene selection was associated simultaneously with the p-value of expression variation and the occurrence rate parameter associated with the percentage of donors whose transcriptomic profile is similar. Our method intertwines stringency level settings, biological data and a knowledge database to highlight molecular interactions using networks and pathways. Analysis performed during iterations helped us to select the optimal threshold required for the most pertinent selection of differentially expressed genes.

Conclusions/significance: We have applied this approach to the well documented mechanism of human macrophage response to lipopolysaccharide stimulation. We thus verified that our method was able to determine with the highest degree of accuracy the best threshold for selecting genes that are truly differentially expressed.

Show MeSH
Interconnections between different networks.From our 195 differentially expressed genes, and the applied parameters (EV = 1.28; EV occurrence≥4/6), the data base has identified 22 different networks. The first 13 networks are heavily inter-connected as shown by solid lines between the networks. The integer beside each line indicates the number of genes that two networks have in common. Networks from 14 to 22 do not share common genes.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC2958130&req=5

pone-0013518-g002: Interconnections between different networks.From our 195 differentially expressed genes, and the applied parameters (EV = 1.28; EV occurrence≥4/6), the data base has identified 22 different networks. The first 13 networks are heavily inter-connected as shown by solid lines between the networks. The integer beside each line indicates the number of genes that two networks have in common. Networks from 14 to 22 do not share common genes.

Mentions: Most differentially expressed genes are grouped within a single meta network (Figure 2). Using EV threshold≥1.28 and EV occurrence≥4/6, three hundred genes having undergone significant expression variation were identified in our experiments. IPA mapped 277 genes, of which 202 could be associated to 22 networks. Thirteen of these are main networks interconnected by at least one common gene, and together form a meta network. The remaining 9 networks are deemed to be independent. For example, the major network #1 shares: a common gene product with network 5, two gene products with network 6, and one with networks 7, 10 and 13, respectively. The composition of each network is given in Table 3, which are classified into major and minor networks according to their score and to the number of genes identified and linked to these networks. The first twelve networks were identified as major networks with fairly high scores ranging from 38 for the best of them to 21 for the 12th. These identified networks are made up of gene products selected according to their EV ratio; varying from 23/35 (66%) for the first network to 15/35 (43%) for the 12th network. It is worth noting that the 13th network shares at least one common gene product with 7 of the major networks. Although it generated a low score (Score = 3), this network strongly overlaps with the other networks and can be considered as a member of the meta network presented in Figure 2. Consequently, the latter is deemed to be made up 13 major networks.


Intertwining threshold settings, biological data and database knowledge to optimize the selection of differentially expressed genes from microarray.

Chuchana P, Holzmuller P, Vezilier F, Berthier D, Chantal I, Severac D, Lemesre JL, Cuny G, Nirdé P, Bucheton B - PLoS ONE (2010)

Interconnections between different networks.From our 195 differentially expressed genes, and the applied parameters (EV = 1.28; EV occurrence≥4/6), the data base has identified 22 different networks. The first 13 networks are heavily inter-connected as shown by solid lines between the networks. The integer beside each line indicates the number of genes that two networks have in common. Networks from 14 to 22 do not share common genes.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC2958130&req=5

pone-0013518-g002: Interconnections between different networks.From our 195 differentially expressed genes, and the applied parameters (EV = 1.28; EV occurrence≥4/6), the data base has identified 22 different networks. The first 13 networks are heavily inter-connected as shown by solid lines between the networks. The integer beside each line indicates the number of genes that two networks have in common. Networks from 14 to 22 do not share common genes.
Mentions: Most differentially expressed genes are grouped within a single meta network (Figure 2). Using EV threshold≥1.28 and EV occurrence≥4/6, three hundred genes having undergone significant expression variation were identified in our experiments. IPA mapped 277 genes, of which 202 could be associated to 22 networks. Thirteen of these are main networks interconnected by at least one common gene, and together form a meta network. The remaining 9 networks are deemed to be independent. For example, the major network #1 shares: a common gene product with network 5, two gene products with network 6, and one with networks 7, 10 and 13, respectively. The composition of each network is given in Table 3, which are classified into major and minor networks according to their score and to the number of genes identified and linked to these networks. The first twelve networks were identified as major networks with fairly high scores ranging from 38 for the best of them to 21 for the 12th. These identified networks are made up of gene products selected according to their EV ratio; varying from 23/35 (66%) for the first network to 15/35 (43%) for the 12th network. It is worth noting that the 13th network shares at least one common gene product with 7 of the major networks. Although it generated a low score (Score = 3), this network strongly overlaps with the other networks and can be considered as a member of the meta network presented in Figure 2. Consequently, the latter is deemed to be made up 13 major networks.

Bottom Line: Analysis performed during iterations helped us to select the optimal threshold required for the most pertinent selection of differentially expressed genes.We have applied this approach to the well documented mechanism of human macrophage response to lipopolysaccharide stimulation.We thus verified that our method was able to determine with the highest degree of accuracy the best threshold for selecting genes that are truly differentially expressed.

View Article: PubMed Central - PubMed

Affiliation: INSERM, Unité 844 - Montpellier, France. paul.chuchana@inserm.fr

ABSTRACT

Background: Many tools used to analyze microarrays in different conditions have been described. However, the integration of deregulated genes within coherent metabolic pathways is lacking. Currently no objective selection criterion based on biological functions exists to determine a threshold demonstrating that a gene is indeed differentially expressed.

Methodology/principal findings: To improve transcriptomic analysis of microarrays, we propose a new statistical approach that takes into account biological parameters. We present an iterative method to optimise the selection of differentially expressed genes in two experimental conditions. The stringency level of gene selection was associated simultaneously with the p-value of expression variation and the occurrence rate parameter associated with the percentage of donors whose transcriptomic profile is similar. Our method intertwines stringency level settings, biological data and a knowledge database to highlight molecular interactions using networks and pathways. Analysis performed during iterations helped us to select the optimal threshold required for the most pertinent selection of differentially expressed genes.

Conclusions/significance: We have applied this approach to the well documented mechanism of human macrophage response to lipopolysaccharide stimulation. We thus verified that our method was able to determine with the highest degree of accuracy the best threshold for selecting genes that are truly differentially expressed.

Show MeSH