Limits...
Assessment of subnetwork detection methods for breast cancer.

Jiang B, Gribskov M - Cancer Inform (2014)

Bottom Line: Here, we compare the results of eight methods: simulated annealing-based jActiveModules, greedy search-based jActiveModules, DEGAS, BioNet, NetBox, ClustEx, OptDis, and NetWalker.While the number of genes/proteins and protein interactions detected by the eight methods vary widely, a core set of 60 genes and 50 interactions was found to be shared by the subnetworks identified by five or more of the methods.Within the core set, 12 genes were found to be known breast cancer genes.

View Article: PubMed Central - PubMed

Affiliation: Department of Biological Sciences, Purdue University, West Lafayette, IN, USA.

ABSTRACT
Subnetwork detection is often used with differential expression analysis to identify modules or pathways associated with a disease or condition. Many computational methods are available for subnetwork analysis. Here, we compare the results of eight methods: simulated annealing-based jActiveModules, greedy search-based jActiveModules, DEGAS, BioNet, NetBox, ClustEx, OptDis, and NetWalker. These methods represent distinctly different computational strategies and are among the most widely used. Each of these methods was used to analyze gene expression data consisting of paired tumor and normal samples from 50 breast cancer patients. While the number of genes/proteins and protein interactions detected by the eight methods vary widely, a core set of 60 genes and 50 interactions was found to be shared by the subnetworks identified by five or more of the methods. Within the core set, 12 genes were found to be known breast cancer genes.

No MeSH data available.


Related in: MedlinePlus

ROC curves of −log10 of the P-values predicting the eight subnetworks. The numbers in the brackets are the AUC.
© Copyright Policy - open-access
Related In: Results  -  Collection


getmorefigures.php?uid=PMC4256043&req=5

f2-cin-suppl.6-2014-015: ROC curves of −log10 of the P-values predicting the eight subnetworks. The numbers in the brackets are the AUC.

Mentions: To further examine the specificity and sensitivity of significant gene coverage of each method, we label each detected gene as a positive sample for each method and examined whether the expression P-values predict the eight subnetworks. We plot eight Receiver Operating Characteristic (ROC) curves in Figure 2 to show the predictability of the P-values for the eight subnetworks. From Figure 2, we find that the top method is BioNet since it achieves an area under the curve (AUC) of 0.93, the highest AUC for any method. This is particularly interesting since BioNet does not depend on a seed gene set. NetBox achieves comparably high AUC (0.89), but there is an obvious kink point on the curve due to the selection of input seed genes based on P-values. The AUC of OptDis ranks the third, probably due to the small size of the subnetwork. jAM.SA detects the largest subnetwork but does not perform well in covering high P-value genes since it accepts a low P-value gene with a specific probability at each iteration in order to avoid suboptimality. ClustEx does not perform as well as NetBox, even though they use the same seed gene set and network data. This is because we only consider the largest subnetwork (210 seeds out of 801 genes) found by ClustEx as the output and discard the smaller subnetworks, which include 455 seeds.


Assessment of subnetwork detection methods for breast cancer.

Jiang B, Gribskov M - Cancer Inform (2014)

ROC curves of −log10 of the P-values predicting the eight subnetworks. The numbers in the brackets are the AUC.
© Copyright Policy - open-access
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC4256043&req=5

f2-cin-suppl.6-2014-015: ROC curves of −log10 of the P-values predicting the eight subnetworks. The numbers in the brackets are the AUC.
Mentions: To further examine the specificity and sensitivity of significant gene coverage of each method, we label each detected gene as a positive sample for each method and examined whether the expression P-values predict the eight subnetworks. We plot eight Receiver Operating Characteristic (ROC) curves in Figure 2 to show the predictability of the P-values for the eight subnetworks. From Figure 2, we find that the top method is BioNet since it achieves an area under the curve (AUC) of 0.93, the highest AUC for any method. This is particularly interesting since BioNet does not depend on a seed gene set. NetBox achieves comparably high AUC (0.89), but there is an obvious kink point on the curve due to the selection of input seed genes based on P-values. The AUC of OptDis ranks the third, probably due to the small size of the subnetwork. jAM.SA detects the largest subnetwork but does not perform well in covering high P-value genes since it accepts a low P-value gene with a specific probability at each iteration in order to avoid suboptimality. ClustEx does not perform as well as NetBox, even though they use the same seed gene set and network data. This is because we only consider the largest subnetwork (210 seeds out of 801 genes) found by ClustEx as the output and discard the smaller subnetworks, which include 455 seeds.

Bottom Line: Here, we compare the results of eight methods: simulated annealing-based jActiveModules, greedy search-based jActiveModules, DEGAS, BioNet, NetBox, ClustEx, OptDis, and NetWalker.While the number of genes/proteins and protein interactions detected by the eight methods vary widely, a core set of 60 genes and 50 interactions was found to be shared by the subnetworks identified by five or more of the methods.Within the core set, 12 genes were found to be known breast cancer genes.

View Article: PubMed Central - PubMed

Affiliation: Department of Biological Sciences, Purdue University, West Lafayette, IN, USA.

ABSTRACT
Subnetwork detection is often used with differential expression analysis to identify modules or pathways associated with a disease or condition. Many computational methods are available for subnetwork analysis. Here, we compare the results of eight methods: simulated annealing-based jActiveModules, greedy search-based jActiveModules, DEGAS, BioNet, NetBox, ClustEx, OptDis, and NetWalker. These methods represent distinctly different computational strategies and are among the most widely used. Each of these methods was used to analyze gene expression data consisting of paired tumor and normal samples from 50 breast cancer patients. While the number of genes/proteins and protein interactions detected by the eight methods vary widely, a core set of 60 genes and 50 interactions was found to be shared by the subnetworks identified by five or more of the methods. Within the core set, 12 genes were found to be known breast cancer genes.

No MeSH data available.


Related in: MedlinePlus