Limits...
Indirect genomic effects on survival from gene expression data.

Ferkingstad E, Frigessi A, Lyng H - Genome Biol. (2008)

Bottom Line: In cancer, genes may have indirect effects on patient survival, mediated through interactions with other genes.We propose a novel methodology to detect and quantify indirect effects from gene expression data.We discover indirect effects through several target genes of transcription factors in cancer microarray data, pointing to genetic interactions that play a significant role in tumor progression.

View Article: PubMed Central - HTML - PubMed

Affiliation: Department of Biostatistics and (sfi) Statistics for Innovation, University of Oslo, Gaustadalleen, Oslo, NO-0314, Norway. egil.ferkingstad@medisin.uio.no

ABSTRACT
In cancer, genes may have indirect effects on patient survival, mediated through interactions with other genes. Methods to study the indirect effects that contribute significantly to survival are not available. We propose a novel methodology to detect and quantify indirect effects from gene expression data. We discover indirect effects through several target genes of transcription factors in cancer microarray data, pointing to genetic interactions that play a significant role in tumor progression.

Show MeSH

Related in: MedlinePlus

Dynamic path models for the Dutch breast cancer data set. The top panel shows the thinned survival forest after selecting genetic interactions for which an indirect and direct effect likely existed. Black arrows indicate a total of 19 significant interactions. The thinned forest consisted of eight networks. A number of dynamic path models were fitted to different sub-networks of these networks: Each connected component, each rooted subtree (that is, each gene with all of its descendants), and each interaction separately. For ten models there was at least one significant indirect effect, indicated with rectangles of different colors. Below the thinned survival forest, the ten models with at least one significant indirect effect are shown. Interactions with significant direct or indirect effects are marked with red arrows. The plus and minus signs on arrows between two genes indicate transcriptional activation and repression, respectively, whereas the plus and minus signs on arrows pointing to survival (dN(t)) indicates that poor survival is associated with activation and repression of the gene, respectively. For each significant path, the average strength of the direct and indirect effect during the first five years is listed, along with a 95% bootstrap confidence interval.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2397510&req=5

Figure 1: Dynamic path models for the Dutch breast cancer data set. The top panel shows the thinned survival forest after selecting genetic interactions for which an indirect and direct effect likely existed. Black arrows indicate a total of 19 significant interactions. The thinned forest consisted of eight networks. A number of dynamic path models were fitted to different sub-networks of these networks: Each connected component, each rooted subtree (that is, each gene with all of its descendants), and each interaction separately. For ten models there was at least one significant indirect effect, indicated with rectangles of different colors. Below the thinned survival forest, the ten models with at least one significant indirect effect are shown. Interactions with significant direct or indirect effects are marked with red arrows. The plus and minus signs on arrows between two genes indicate transcriptional activation and repression, respectively, whereas the plus and minus signs on arrows pointing to survival (dN(t)) indicates that poor survival is associated with activation and repression of the gene, respectively. For each significant path, the average strength of the direct and indirect effect during the first five years is listed, along with a 95% bootstrap confidence interval.

Mentions: First, we illustrate the results that are obtained with our method, using the genes PPARD (encoding peroxisome proliferator-activated receptor D) and ADFP (encoding adipose differentiation-related protein) as an example (Figure 1, model 2). All details are explained in the subsequent text. We have gene expression data for both genes from cancer patients and censored survival data from the same patients. It is known that expression of PPARD influences expression of ADFP. An effect of PPARD on survival could, therefore, be mediated through ADFP. In our terminology, this is an indirect effect of PPARD on survival, through ADFP. Other indirect effects of PPARD, through other genes, could also exist, and PPARD could also have a direct effect on survival, that is, an effect that is not mediated through any other genes in our data set. Using our method, we can discover and quantify the strengths of such indirect and direct effects. Specifically, we found that, summed over the first five years, PPARD had a direct effect on survival of 0.141 (with a 95% bootstrap confidence interval of (0.047, 0.206)), and an indirect effect of 0.048 (95% confidence interval of (0.030, 0.101)). In this case, all effects are positive, indicated by plus signs on the arrows in Figure 1. In other cases, the effects can be negative, indicated by minus signs. Positive effects are harmful (increase the risk of death), while negative effects are beneficial. Since the bootstrap confidence intervals do not contain zero, both the direct and indirect effects are significant. The 'total effect' is simply the sum of the direct and indirect effects. Here, approximately 24% of the total effect is indirect.


Indirect genomic effects on survival from gene expression data.

Ferkingstad E, Frigessi A, Lyng H - Genome Biol. (2008)

Dynamic path models for the Dutch breast cancer data set. The top panel shows the thinned survival forest after selecting genetic interactions for which an indirect and direct effect likely existed. Black arrows indicate a total of 19 significant interactions. The thinned forest consisted of eight networks. A number of dynamic path models were fitted to different sub-networks of these networks: Each connected component, each rooted subtree (that is, each gene with all of its descendants), and each interaction separately. For ten models there was at least one significant indirect effect, indicated with rectangles of different colors. Below the thinned survival forest, the ten models with at least one significant indirect effect are shown. Interactions with significant direct or indirect effects are marked with red arrows. The plus and minus signs on arrows between two genes indicate transcriptional activation and repression, respectively, whereas the plus and minus signs on arrows pointing to survival (dN(t)) indicates that poor survival is associated with activation and repression of the gene, respectively. For each significant path, the average strength of the direct and indirect effect during the first five years is listed, along with a 95% bootstrap confidence interval.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2397510&req=5

Figure 1: Dynamic path models for the Dutch breast cancer data set. The top panel shows the thinned survival forest after selecting genetic interactions for which an indirect and direct effect likely existed. Black arrows indicate a total of 19 significant interactions. The thinned forest consisted of eight networks. A number of dynamic path models were fitted to different sub-networks of these networks: Each connected component, each rooted subtree (that is, each gene with all of its descendants), and each interaction separately. For ten models there was at least one significant indirect effect, indicated with rectangles of different colors. Below the thinned survival forest, the ten models with at least one significant indirect effect are shown. Interactions with significant direct or indirect effects are marked with red arrows. The plus and minus signs on arrows between two genes indicate transcriptional activation and repression, respectively, whereas the plus and minus signs on arrows pointing to survival (dN(t)) indicates that poor survival is associated with activation and repression of the gene, respectively. For each significant path, the average strength of the direct and indirect effect during the first five years is listed, along with a 95% bootstrap confidence interval.
Mentions: First, we illustrate the results that are obtained with our method, using the genes PPARD (encoding peroxisome proliferator-activated receptor D) and ADFP (encoding adipose differentiation-related protein) as an example (Figure 1, model 2). All details are explained in the subsequent text. We have gene expression data for both genes from cancer patients and censored survival data from the same patients. It is known that expression of PPARD influences expression of ADFP. An effect of PPARD on survival could, therefore, be mediated through ADFP. In our terminology, this is an indirect effect of PPARD on survival, through ADFP. Other indirect effects of PPARD, through other genes, could also exist, and PPARD could also have a direct effect on survival, that is, an effect that is not mediated through any other genes in our data set. Using our method, we can discover and quantify the strengths of such indirect and direct effects. Specifically, we found that, summed over the first five years, PPARD had a direct effect on survival of 0.141 (with a 95% bootstrap confidence interval of (0.047, 0.206)), and an indirect effect of 0.048 (95% confidence interval of (0.030, 0.101)). In this case, all effects are positive, indicated by plus signs on the arrows in Figure 1. In other cases, the effects can be negative, indicated by minus signs. Positive effects are harmful (increase the risk of death), while negative effects are beneficial. Since the bootstrap confidence intervals do not contain zero, both the direct and indirect effects are significant. The 'total effect' is simply the sum of the direct and indirect effects. Here, approximately 24% of the total effect is indirect.

Bottom Line: In cancer, genes may have indirect effects on patient survival, mediated through interactions with other genes.We propose a novel methodology to detect and quantify indirect effects from gene expression data.We discover indirect effects through several target genes of transcription factors in cancer microarray data, pointing to genetic interactions that play a significant role in tumor progression.

View Article: PubMed Central - HTML - PubMed

Affiliation: Department of Biostatistics and (sfi) Statistics for Innovation, University of Oslo, Gaustadalleen, Oslo, NO-0314, Norway. egil.ferkingstad@medisin.uio.no

ABSTRACT
In cancer, genes may have indirect effects on patient survival, mediated through interactions with other genes. Methods to study the indirect effects that contribute significantly to survival are not available. We propose a novel methodology to detect and quantify indirect effects from gene expression data. We discover indirect effects through several target genes of transcription factors in cancer microarray data, pointing to genetic interactions that play a significant role in tumor progression.

Show MeSH
Related in: MedlinePlus