Gene Co-expression Network Construction

After MAS normalization of all Affymetrix microarray samples, outliers were detected using the arrayQualityMetricsBioconductor package, which uses three different statistical tests to identify outliers (Gentleman et al.[2004 (link)]; Kauffmann et al.[2009 (link)]). Seventy-four samples failed at least one test and were considered as outliers and removed from the dataset. As a result, a total of 1,081 Affymetrix samples remained for co-expression analysis. For genes with multiple Affymetrixprobesets matched, the probeset with highest expression profile was used. There are several kinds of methods to evaluate the strength of co-expression, such as Pearson correlation coefficient (PCC), mutual rank (MR) based on rank transformations of the weighted PCC (Obayashi and Kinoshita[2009 (link)]) and correspondence analysis (CA) (Yano et al.[2006 (link)]). Although PCC takes a long-calculation time and was considered to contain many false-positives (Hamada et al.[2011 (link)]; Obayashi et al.[2011 (link)]), it has been widely used as an index in the co-expression analysis, such as RiceArrayNet (Lee et al.[2009 (link)]), RiceXPro (Sato et al.[2011 (link)]) and Gene Co-expression Network Browser (Ficklin et al.[2010 (link)]). The success in functional study of plant genes using PCC has also been reported (Fujii et al.[2010 (link)]; Matsuura et al.[2010 (link)]; Soeno et al.[2010 (link)]). Therefore, we adopted PCC to measure tendency of co-expression between genes based on these 1,081 Affymetrix samples. To choose an appropriate PCC cutoff value to construct co-expression network, we examined the changes in the node number, edge number, and network density as a function of PCC cutoff values. As the cutoff value increased, both the node number and edge number decreased; however, as the cutoff reached a relatively high value, the decreasing rate of edges became slower than that of nodes, which might lead to an increase in the network density. Indeed, the network density showed minima around 0.75 (general) and 0.8 (abiotic and biotic stresses) PCC cutoff values and increased thereafter. Therefore, we selected default values of the PCC cutoff as 0.75 and 0.8, for general and abiotic and biotic stresses, respectively. Cytoscape Web, an interactive web-based network browser, was used as the network viewer (Lopes et al.[2010 (link)]).