Differential Network Entropy Reveals Cancer System HallmarksJAKE PHILLIPS, LINH HUYNH

Network Modeling• Node – represents a measured

parameter (i.e. gene or protein expression)

• Edge – represents a significant correlation between two node elements

Figure 1. Protein Gene Network pathwaycommons.org

Differential Networks• Describe changes

between expression and interaction data

• Includes more information than networks describing solely differences in gene expression

Figure 2. Differential Network Design (3)

Network Entropy

Figure 3. Illustrating Entropy in Networks (1)

• Network entropy is a measure of the unpredictability of response as modeled by a network

• Average shortest path length describes the mean separation between node elements

• Highly entropic systems have shorter average path lengths and are more robust to node deletion

Abstract• Cellular phenotype can be

modeled as a network describing expression and interaction of genes

• This network can be analyzed to reveal characteristics which differentiate healthy and diseased cells amongst other properties

• Entropy analysis at global and local levels provides insight into how these systems can be used to identify novel drug targets.

Local Network

Global Network

Results – Cancer vs Normal

Figure 4. Local Network Entropy, Cancer vs Normal (2)

• Cancer and Normal cells are considered to have the same nodes and interactions, only varying in their weights

Results – Global vs Local

• Networks modeling cancer cells have greater local and global entropy than those for normal cells

• Local network entropy is a better determinant of a gene’s potential as a biomarker than global network entropy

Results – Activated vs Inactivated Cancer Genes• Differential network entropy and differential expression are anti-

correlated

Inactivated Cancer Gene

Lower Expressio

n

Increased Local

Entropy

Activated Cancer Gene

Increased Expressio

n

Decreased Local

Entropy

Therapeutic Implication

• Identification of biomarkers and therapeutic targets• Genes expressed more in cancer are likely to have a

significant reduction in network entropy• Allows for oncogene identification by observation of

decreases in local network entropy

References

(1)Lloyd Demetrius, Thomas Manke, Robustness and network evolution—an entropic principle, Physica A: Statistical Mechanics and its Applications, Volume 346, Issues 3–4, 15 February 2005, Pages 682-696

(2)West, James et al. “Differential Network Entropy Reveals Cancer System Hallmarks.” Scientific Reports 2 (2012): 802. PMC. Web. 28 July 2015.

(3)Zeng T, Sun SY, Wang Y, et al. Network biomarkers reveal dysfunctional gene regulations during disease progression. FEBS J.2013; 280: 5682–95.