Limits...
Uncover disease genes by maximizing information flow in the phenome-interactome network.

Chen Y, Jiang T, Jiang R - Bioinformatics (2011)

Bottom Line: We demonstrate the effectiveness of this method in prioritizing candidate genes through a series of cross-validation experiments, and we show the possibility of using this method to identify diseases with which a query gene may be associated.As an example application, we apply our method to predict driver genes in 50 copy number aberration regions of melanoma.Our method is not only able to identify several driver genes that have been reported in the literature, it also shed some new biological insights on the understanding of the modular property and transcriptional regulation scheme of these driver genes. ruijiang@tsinghua.edu.cn.

View Article: PubMed Central - PubMed

Affiliation: MOE Key Laboratory of Bioinformatics and Bioinformatics Division, TNLIST/Department of Automation, Tsinghua University, Beijing 1000084, China.

ABSTRACT

Motivation: Pinpointing genes that underlie human inherited diseases among candidate genes in susceptibility genetic regions is the primary step towards the understanding of pathogenesis of diseases. Although several probabilistic models have been proposed to prioritize candidate genes using phenotype similarities and protein-protein interactions, no combinatorial approaches have been proposed in the literature.

Results: We propose the first combinatorial approach for prioritizing candidate genes. We first construct a phenome-interactome network by integrating the given phenotype similarity profile, protein-protein interaction network and associations between diseases and genes. Then, we introduce a computational method called MAXIF to maximize the information flow in this network for uncovering genes that underlie diseases. We demonstrate the effectiveness of this method in prioritizing candidate genes through a series of cross-validation experiments, and we show the possibility of using this method to identify diseases with which a query gene may be associated. We demonstrate the competitive performance of our method through a comparison with two existing state-of-the-art methods, and we analyze the robustness of our method with respect to the parameters involved. As an example application, we apply our method to predict driver genes in 50 copy number aberration regions of melanoma. Our method is not only able to identify several driver genes that have been reported in the literature, it also shed some new biological insights on the understanding of the modular property and transcriptional regulation scheme of these driver genes.

Contact: ruijiang@tsinghua.edu.cn.

Show MeSH

Related in: MedlinePlus

Comparison with existing methods on leave-one-out cross-validation experiments against random genes. (A) The ROC curve. (B) The precision-recall curve.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3117332&req=5

Figure 3: Comparison with existing methods on leave-one-out cross-validation experiments against random genes. (A) The ROC curve. (B) The precision-recall curve.

Mentions: We evaluate the performance using the AUC score and present the results in Figure 3A. The figure shows that the ROC curve of MAXIF lies clearly above those of PRINCE and RWRH. More specifically, the AUC scores are 90.76% for MAXIF, 86.84% for PRINCE and 86.94% for RWRH. In addition, the PRE values are 56.89% for MAXIF, 50.02% for PRINCE and 50.18% for RWRH and the MRR values are 10.03% for MAXIF, 13.80% for PRINCE and 13.81% for RWRH.Fig. 3.


Uncover disease genes by maximizing information flow in the phenome-interactome network.

Chen Y, Jiang T, Jiang R - Bioinformatics (2011)

Comparison with existing methods on leave-one-out cross-validation experiments against random genes. (A) The ROC curve. (B) The precision-recall curve.
© Copyright Policy - creative-commons
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3117332&req=5

Figure 3: Comparison with existing methods on leave-one-out cross-validation experiments against random genes. (A) The ROC curve. (B) The precision-recall curve.
Mentions: We evaluate the performance using the AUC score and present the results in Figure 3A. The figure shows that the ROC curve of MAXIF lies clearly above those of PRINCE and RWRH. More specifically, the AUC scores are 90.76% for MAXIF, 86.84% for PRINCE and 86.94% for RWRH. In addition, the PRE values are 56.89% for MAXIF, 50.02% for PRINCE and 50.18% for RWRH and the MRR values are 10.03% for MAXIF, 13.80% for PRINCE and 13.81% for RWRH.Fig. 3.

Bottom Line: We demonstrate the effectiveness of this method in prioritizing candidate genes through a series of cross-validation experiments, and we show the possibility of using this method to identify diseases with which a query gene may be associated.As an example application, we apply our method to predict driver genes in 50 copy number aberration regions of melanoma.Our method is not only able to identify several driver genes that have been reported in the literature, it also shed some new biological insights on the understanding of the modular property and transcriptional regulation scheme of these driver genes. ruijiang@tsinghua.edu.cn.

View Article: PubMed Central - PubMed

Affiliation: MOE Key Laboratory of Bioinformatics and Bioinformatics Division, TNLIST/Department of Automation, Tsinghua University, Beijing 1000084, China.

ABSTRACT

Motivation: Pinpointing genes that underlie human inherited diseases among candidate genes in susceptibility genetic regions is the primary step towards the understanding of pathogenesis of diseases. Although several probabilistic models have been proposed to prioritize candidate genes using phenotype similarities and protein-protein interactions, no combinatorial approaches have been proposed in the literature.

Results: We propose the first combinatorial approach for prioritizing candidate genes. We first construct a phenome-interactome network by integrating the given phenotype similarity profile, protein-protein interaction network and associations between diseases and genes. Then, we introduce a computational method called MAXIF to maximize the information flow in this network for uncovering genes that underlie diseases. We demonstrate the effectiveness of this method in prioritizing candidate genes through a series of cross-validation experiments, and we show the possibility of using this method to identify diseases with which a query gene may be associated. We demonstrate the competitive performance of our method through a comparison with two existing state-of-the-art methods, and we analyze the robustness of our method with respect to the parameters involved. As an example application, we apply our method to predict driver genes in 50 copy number aberration regions of melanoma. Our method is not only able to identify several driver genes that have been reported in the literature, it also shed some new biological insights on the understanding of the modular property and transcriptional regulation scheme of these driver genes.

Contact: ruijiang@tsinghua.edu.cn.

Show MeSH
Related in: MedlinePlus