Transmembrane alpha helix predictions
Up to 18 individual programs contribute to the prediction of transmembrane alpha helices (see Resource > TM spans (alpha helix) for prediction program details). The programs PHDhtm, MemSat_v3 and Minnou take into account homologous proteins for their predictions. For PHDhtm the alignments and the final prediction are done by Network Sequence Protein Analysis (Lyon). If there are no homologous proteins available at the SwissProt database PHDhtm runs in single sequence mode (as all other prediction programs used).
Consensus TM alpha helix prediction (AramTmCon)
The individual predictions for each protein are combined to a built-in consensus prediction using the Bayes' theorem. A consensus score is based on the ratio of the posterior probabilities of both hypotheses, i.e. that a certain sequence is a TM region given the program predicts positively and that a certain sequence is not a TM region although the program predicts positively. The corresponding likelihoods of the hypotheses were estimated using a set of training proteins with known TM segments.
The consensus score value is normalised to a maximal value of 1. (Note that a score value of 1 does not reflect a probability of 1.) The consensus diagram shows TM segments with a score above 0.10 and counts all with a score equal to or above 0.42. The center and edge positions for each consensus TM segment are calculated by the average positions. Extra procedures are added to avoid overlapping or unreasonably short consensus segments.
An additional consensus prediction (performed by ConPred_v2) is available for many of the proteins. Notice that ConPred_v2 uses a different set of individual predictions.
Extendend consensus TM alpha helix prediction (AramTmMultiCon)
A built-in extended consensus prediction is performed by combining consensus predictions of several homologous proteins. The algorithm maps TM alpha helix consensus predictions to a multiple alignment of homologous protein sequences, and projects the resulting new consensus to the protein sequence in question. The consensus score value is normalised to a maximal value of 1. (As for the single consensus prediction a score value of 1 does not reflect a probability of 1.) The consensus diagram shows TM segments with a score above 0.10 and counts all with a score equal to or above 0.38.
Transmembrane alpha helix details
The following details of the predicted alpha helices are available
- positions on protein
- mean hydrophobicities
- maximal amphiphilicities
- hydropathy profile
The position data are taken from the output of the TM prediction programs. The calculation of the average hydrophobicities are based on the hydrophobicity scale published by Eisenberg et al. (1984). The calculation of the maximal amphiphilicities assumes 3.7 amino acids per cycle period for a typical alpha-helix. As method for detecting periodicity we use a Fourier transform and calculate the amphiphilicity as maximal sum of Fourier intensities of a 12-residue segment within a predicted TM region (Eisenberg et al. (1984); Cornette et al. (1987)). The hydropathy profile feature has been adapted from John Ward (University of Minnesota, St. Paul) who kindly provided the algorithm. The algorithm was slightly modified and is based on the same hydrophobicity scale as used for the average hydrophobicity calculations (see above).
Transmembrane beta barrel prediction
Up to 5 individual programs contribute to the prediction of transmembrane beta barrel proteins. (see Resource > TM spans (beta barrel) for prediction program details).