Poster Sessions

BIOINFO-15

Jeff Skinner

J. Skinner, S. Varma, A. Yambartsev, Y. Kotliarov, Y. Huyen, A. Morgun

Novel Tool for Identification and Comparison of Gene-Gene Association Networks

Networks of co-expressed genes closely model real biological systems. Comparisons of co-expression networks may reveal systems level trends, like splitting or merging of networks. We developed a novel tool to compute Pearson, Spearman and Kendall correlations for each gene-gene pair and statistically compare these associations between two classes. Individual gene-gene associations define edge weights in co-expression networks, while statistical tests determine presence or absence of network edges between classes. We evaluated the sensitivity and specificity of all three correlations with simulated microarray data for several sample sizes and differences in correlation using ROC curves. We present results from 250 simulation runs using data with 40 genes per chip. With fixed difference in correlation from r = -0.5 to +0.5, simulations with n = 6 chips per class produced ROC curves with 72.49-76.61% area under curve (AUC) and n = 15 chips per class produced 90.38-93.35% AUC. With fixed sample size n = 5 chips per class, simulations with r = -0.65 to +0.65 produced 74.59-76.47% AUC and r = -0.85 to +0.85 produced 93.09-96.46% AUC. We have developed a promising new tool for analysis of gene co-expression networks that allows for robust comparison of networks between two classes.