Fig. 3
From: iREAD: a tool for intron retention detection from RNA-seq data
Comparison of the precision of iREAD with IRFinder on their identified intron retention events at default settings using simulated RNA-seq data with known ground-truth intron retention events. Two datasets were used, which are, for convenience, called15M dataset (containing 15 M reads) and 30 M dataset (containing 30 M reads). Firstly, retained introns were detected using both IRFinder and iREAD with default thresholds. Suppose N1 and N2 IR events were detected by each method, and the top ranked N = min(N1, N2) IR events of each method were compared in terms of the number of shared IR events and precision. N was found to be 2498 and 2683 for the 15 M (upper panel) and 30 M (lower panel) dataset, respectively. The sharing of identified IRs on the 15 M and 30 M dataset is shown in (a) and (e), respectively. On the 15 M dataset, the precision of IRFinder and iREAD based on the GS1 (FPKM≥0.1), GS2 (FPKM≥0.3) and GS3 (FPKM≥0.5) gold standard is presented in (b), (c) and (d), respectively. On the 30 M dataset, the precision of IRFinder and iREAD based on the GS1, GS2 and GS3 gold standard is presented in (f), (g) and (h), respectively. The results showed that iREAD achieved consistently higher precision than IRFinder on the IRs identified at default setting