Skip to main content

Table 3 Comparison of MDL, best-fit-I, and best-fit-II with CMI- and MDL-based filtering for yeast-pathway data

From: Using the minimum description length principle to reduce the rate of false positives of best-fit algorithms

Algorithm Noise = 0 Noise = 5% Noise = 10%
TP FP μ ham e μssd TP FP μ ham e μssd TP FP μ ham e μssd
MDL 14 2 0.65 1.31 11.5 9 0.93 1.42 8.9 12.5 1.11 1.45
BF-I 15 5 0.71 1.25 12.2 11.9 0.99 1.44 9.8 18.4 1.25 1.49
BF-I-CMI 11 1 0.71 1.43 10.4 9 0.96 1.47 8.3 14 1.17 1.51
BF-I-MDL 14 2 0.65 1.17 10.8 8.5 0.93 1.43 8.6 13.1 1.13 1.48
BF-II 15 3 0.65 1.41 12.4 10.4 0.94 1.45 10.6 16.5 1.17 1.48
BF-II-CMI 12 2 0.71 1.46 11 8.7 0.93 1.47 8.3 12.4 1.12 1.50
BF-II-MDL 13 1 0.65 1.36 11.1 7.7 0.9 1.42 9.2 11.9 1.08 1.44