Valid and invalid implementations of GOR secondary structure predictions

GOR algorithms have long been a standard methodology for predicting protein secondary structure from primary sequence. We have developed two short validation sequences for the GOR I and GOR II algorithms. Use of these sequences with seven commercial and non-commercial implementations of these algorithms demonstrated that several were incorrect implementations, including two of the three commercial modules implementing the GOR I algorithm. This may be due to an easy misinterpretation of the GOR I algorithm and related data tables. We present the validation sequences and discuss implications of this widely propagated error on secondary and tertiary structure prediction, using several proteins of known structure in three different structural classes as examples. A valid GOR I implementation predicts secondary structure increases the accuracy of predictions by from 1-13 percentage points over an invalid implementation based on the easy misinterpretation. A valid implementation of the GOR I and GOR II algorithms is available from the authors.