Figure 5
From: Similarities and differences of polyadenylation signals in human and fly
The double model including the deduced DSE matrices improves the discrimination of real 3'UTRs from control sequences. ROC curves showing comparison of best scores (Forward decoding). True positive rate was computed for 3'880 3'000 bp long human sequences (A) and 840 600 bp long Drosophila sequences (B). False positive rate was computed on the test set sequences randomized in 100 bases windows. A-B: Double model (black curve) is more specific than PAS-only model (red), and that ERPIN 3.1 predictions. The DSE matrix included in the double model allows better recognition than a U-rich 25 nt matrix (green) or modified versions of the DSE, where position 4 or 6 (human) or 2 (Drosophila) have been replaced with background nucleotide frequency.