Table 1.

Results obtained from the various predictors, on a dataset of 72 transmembrane proteins [38]. Results obtained when the methods were not trained and tested on the same dataset, however some of the proteins in the dataset were present in the datasets used for training the other methods. The results of HMM-TM were obtained through a nine-fold cross validation procedure. The methods that allow the incorporation of experimental information are listed separately. The results of UMDHMM^TMHP could not be obtained by cross-validation (since it was trained on the same dataset), and thus are listed separately in the text

Method	Q	C	SOV	Correctly predicted TM segments (%)	Correctly predicted Topologies (%)
Methods that allow the incorporation of experimental information
HMM-TM (cross-validation)	0.903	0.762	0.939	59/72 (81.9%)	55/72 (76.4%)
TMHMM	0.902	0.762	0.931	58/72 (80.6%)	49/72 (68.1%)
HMMTOP	0.890	0.735	0.932	58/72 (80.6%)	49/72 (68.1%)
Phobius †	0.911	0.785	0.954	65/72 (90.3%)	52/72 (72.2%)
Methods that do not allow the incorporation of experimental information
MEMSAT	0.905	0.767	0.954	63/72 (87.5%)	48/72 (66.7%)
S-TMHMM †	0.897	0.747	0.925	59/72 (81.9%)	52/72 (72.2%)
PRO-TMHMM* †	0.910	0.779	0.945	65/72 (90.3%)	63/72 (87.5%)
PRODIV-TMHMM* †	0.914	0.794	0.970	67/72 (93.1%)	64/72 (87.5%)

* The methods using evolutionary information are denoted with an asterisk.

† These predictors were trained on sets containing sequences similar to the ones included in the training set we used here