Skip to main content

Table 2 Performance of benchmarking detection methods for holo proteins in terms of: (d) distance (FN) false negatives to PDBsum ground-truth cavity centers; (TP) true positives; (FP) false positives; (TN) true negatives; (S v ) sensitivity; (S c ) specificity; (a) accuracy; (r d ) ratio of detected ground-truth cavities; and (C u ) cumulative number of undetected ground-truth cavities

From: GPU-based detection of protein cavities using Gaussian surfaces

d GaussianFinder ConCavity POCASA SURFNET PASS GHECOM Fpocket 3V KVFinder
d[0.0,1.0] 16081 3133 5234 3668 11738 10438 14063 9174 12493
d]1.0,2.0] 366 239 338 2574 2100 571 432 703 719
d]2.0,3.0] 410 296 419 1063 281 789 406 811 813
d]3.0,4.0] 334 488 609 925 360 932 504 349 418
TP 17191 4156 6600 8230 14479 12730 15405 12049 14443
FP 2460 2155 2083 1806 634 1278 2151 1564 1941
TN 3231 916 2673 1559 4080 4968 3423 2476 2725
FN 440 362 214 227 207 658 391 511 553
S v 0.975 0.919 0.969 0.973 0.986 0.951 0.975 0.959 0.963
S c 0.568 0.298 0.562 0.463 0.866 0.795 0.614 0.612 0.584
a 0.876 0.668 0.801 0.828 0.957 0.901 0.881 0.875 0.873
r d 0.963 0.233 0.369 0.461 0.811 0.713 0.863 0.675 0.809
C u 659 13694 11250 9620 3371 5120 2445 5801 3407