Skip to main content
Figure 6 | BMC Bioinformatics

Figure 6

From: Unsupervised reduction of random noise in complex data by a row-specific, sorted principal component-guided method

Figure 6

Preferential correlation changes between gene pairs. RSPR-NR was applied to the Arabidopsis expression profile data set. The cosine correlations between each of 80,000 randomly sampled gene pairs before and after RSPR-NR were recorded. The angle corresponding to each cosine correlation value was used in the plots. (A) The angle change from before RSPR-NR to after RSPR-NR is plotted along the angle before RSPR-NR, for the sampled gene pairs. A loess fit curve for the distribution is shown in gray. (B) From (A), the loess fit curve values were made zero (gray line) to normalize the distribution in small ranges of the angle before. The angle change value after this loess-based normalization is called the relative angle change value. This relative-angle-change-vs.-angle-before plot is used in (C) and (D). (C) The gene pairs that share GO process and component terms (Jaccard similarity = 0.5) are plotted. (D) The gene pairs that do not share any GO process and component terms and that have more than 5 GO terms in union among the sampled gene pairs are plotted. In both (C) and (D), the gene pairs with low correlation before RSPR-NR (an angle range between 0.3Ï€ and 0.7Ï€) are excluded. The parameters used for RSPR-NR were a subset row number of 200, a repeat number of 20, and an FDR of 0.0316. The dot sizes in the plots were adjusted according to the density of dots in each plot.

Back to article page