Skip to main content

Table 2 Benchmark results for the SVM-based method ('MEMSAT-SVM') against a selection of leading topology predictors

From: Transmembrane protein topology prediction using support vector machines

Method Algorithm Correct helix count Correct helix locations Correct N-terminal FP helix FN helix Correct SP topology Correct RE topology Correct topology
MEMSAT-SVM SVM 95% 91% 91% 4% 5% 93% 64% 89%
OCTOPUS NN + HMM 86% 83% 84% 14% 2% 21% 73% 79%
MEMSAT3 NN 84% 76% 84% 8% 8% 57% 64% 76%
ENSEMBLE NN + HMM 77% 76% 79% 18% 5% 7% 55% 67%
PHOBIUS HMM 75% 76% 79% 9% 16% 93% 36% 63%
HMMTOP HMM 77% 76% 78% 18% 6% 29% 64% 63%
PRODIV HMM 79% 64% 76% 19% 8% 0% 18% 57%
SVMTOP SVM 66% 64% 66% 22% 22% 0% 55% 53%
TMHMM HMM 75% 68% 72% 14% 20% 29% 55% 53%
PHDhtm NN 75% 54% 55% 23% 30% 29% 18% 45%
  1. Column 1: Method – Prediction method. Column 2: Algorithm – Underlying machine-learning algorithm. Column 3: Correct helix count – Fraction of sequences with the correct number of TM helices predicted. Column 4: Correct helix locations – Fraction of sequences with the correct number and locations of TM helices predicted. Column 5: Correct N-terminal – Fraction of sequences with the correct N-terminal location predicted. Column 6: FP helix – Fraction of sequences with at least one over predicted TM helix. Column 7: FN helix – Fraction of sequences with at least one under predicted TM helix. Column 8: Correct SP topology: Fraction of sequences that contain signal peptides that have correct overall topology predicted. Column 9: Correct RE topology: Fraction of sequences that contain re-entrant helices that have correct overall topology predicted. Column 10: Correct topology: Fraction of sequences that have correct overall topology predicted, requiring the correct number and location of TM helices and correct location of the N-terminal. TM helices must overlap their defined positions by at least 5 residues.