Skip to main content
. 2011 Oct 22;8:39. doi: 10.1186/1742-4682-8-39

Table 8.

Performance comparison of the biclustering network using new evaluation criteria.

Methods EdgeC ount TP FP TN FN FP to TP TN to FN AURO C AUPR
Gold 2194 2194 0 400396 0 0 0 1 1

ALL 5440 94 5346 395050 2100 4623 2150 0.7530 0.4614

SAMBA 1611 46 1565 398831 2148 1340 2501 0.6958 0.3644

ISA 2558 56 2502 397894 2138 2141 1151 0.8451 0.6306

OPSM 220 12 208 400188 2182 190 2453 0.5423 0.089

Friedman 947 22 925 399471 2172 794 2700 0.6364 0.2491

CMSBE 735 20 715 399681 2174 653 2750 0.6181 0.2333

K-means 380 13 367 400029 2181 323 3100 0.5667 0.1307

Bivisu 1515 13 1502 398894 2181 1265 2610 0.6845 0.3326

CC 590 3 587 399809 2191 507 2800 0.5943 0.1788

The performances of the various biclustering algorithms improved when false positive edges could be considered true positive edges on the basis of strong evidence in the gold network. Column 7 shows the number of false positive edges in each algorithm that could be considered true positive edges (i.e. have evidence in the gold standard network). Column 8 shows the number of true negative edges that are considered as false negative.