Table 1.
Overview of performance of several methods and measures for prediction of motif enrichment on artificial and real datasets
|
Method or measure |
Type |
Artificial data |
Real data |
Reference |
||||
|---|---|---|---|---|---|---|---|---|
| Recall | Precision | F-measure | Recall | Precision | F-measure | |||
| LocaMo Finder (Gaussian) |
local |
0.755 |
0.609 |
0.674 |
0.371 |
0.757 |
0.498 |
this study |
| LocaMo Finder (uniform) |
local |
0.727 |
0.519 |
0.606 |
0.343 |
0.723 |
0.465 |
this study |
| RSAT (Binomial distribution) ($) |
global |
0.714 |
0.285 |
0.408 |
0.429 |
0.440 |
0.434 |
RSAT [40] |
| ORI (**) |
global |
0.677 |
0.386 |
0.492 |
0.343 |
0.563 |
0.426 |
this study |
| Hypergeometric distribution (*) |
global |
0.745 |
0.272 |
0.399 |
0.400 |
0.450 |
0.424 |
AlignACE [41] |
| Fisher’s exact test (*) |
global |
0.747 |
0.276 |
0.403 |
0.400 |
0.443 |
0.420 |
oPOSSUM [42] |
| ORI (*) |
global |
0.768 |
0.258 |
0.387 |
0.429 |
0.407 |
0.417 |
this study |
| RSAT (Binomial distribution) ($$) |
global |
0.591 |
0.498 |
0.541 |
0.271 |
0.607 |
0.375 |
RSAT [40] |
| Hypergeometric distribution (***) |
global |
0.605 |
0.522 |
0.560 |
0.243 |
0.706 |
0.361 |
AlignACE [41] |
| Fisher’s exact test (***) |
global |
0.605 |
0.530 |
0.565 |
0.243 |
0.667 |
0.356 |
oPOSSUM [42] |
| Casimiro et al. |
local |
0.727 |
0.053 |
0.099 |
0.629 |
0.132 |
0.218 |
[9] |
| Berendzen et al. |
local |
0.859 |
0.044 |
0.083 |
0.786 |
0.093 |
0.167 |
[1] |
| Vardhanabhuti et al. |
local |
0.409 |
0.079 |
0.133 |
0.314 |
0.090 |
0.139 |
[3] |
| FIRE (Information content) |
global |
0.586 |
0.342 |
0.432 |
0.100 |
0.200 |
0.133 |
FIRE [43] |
| TFM-Explorer |
local |
0.432 |
0.145 |
0.217 |
0.186 |
0.076 |
0.108 |
[6] |
| FREE |
local |
0.155 |
0.182 |
0.167 |
0.029 |
0.013 |
0.018 |
[5] |
| A-GLAM | local | 0.032 | 0.259 | 0.057 | 0.000 | 0.000 | NA | [4,27] |
For each method or measure the type of measure (“local”: local enrichment of positioning; “global”: global enrichment), the recall, precision, and F-measure is given for the artificial and real datasets, as well as a reference. Methods are sorted by decreasing F-measure obtained on the real datasets. (*) P value threshold 0.01; (**) P value threshold 0.001; (***) P value threshold 1e-4; ($) sig threshold 0; ($$) sig threshold 2.