The observed Q3 accuracy of ZPRED and MULPRED was between 3 and 8% lower
than the other methods, so a consensus was calculated only from DSC,
PHD, PREDATOR and NNSSP. The standard consensus was calculated by examining
the prediction for each method, at each position and taking the most
popular state. For example if a residue had the following
NNSSP = Helix, PREDATOR = Helix, DSC = Strand, PHD = Helix
the consensus prediction would be Helix. If there was no consensus for a particular residue, the result from the PHD method was used. More complex methods for combining the different predictions were investigated and tested, as discussed in the results section.