Skip to main content
Figure 4 | BMC Bioinformatics

Figure 4

From: In Silico Evaluation of Predicted Regulatory Interactions in Arabidopsis thaliana

Figure 4

Performance of Correlation and CRE Prediction Methods across microarray data: All Expressed Genes (Whole Genome Prediction Analysis). The overall pattern for the PPV (A), Sensitivity (B) and F-Measure (C), Specificity (D), and NPV (E) for the analysis of all genes represented on the ATH1 microarray chip using correlation and/or CRE detection prediction methods for determining TF targets. The results of this analysis show a more highly variable PPV (A) for both CRE detection methods and a consistent PPV for all methods involving the Spearman correlation. Also shown is a reduced and far more variable F measure for the ≥ 1 CRE method indicating a loss of consistently strong performance in balance between the sensitivity and PPV. This loss of performance is also seen in the consistently low specificity (D). The NPV (E) shows a consistent pattern of response across all measures. One interesting observation is that the Spearman correlation combined with CRE over representation has the highest and most consistent PPV and specificity across all datasets for this analysis, comparable to what was seen in the previous analysis (Figure 2). CRE Over: CRE Over-representation, ≥1 CRE: detection of 1 or more CREs, Spearman: Spearman Rank Correlation.

Back to article page