Inverse sequence similarity in proteins and its relation to the three-dimensional fold.

Nowadays the most successful strategy for the prediction of the tertiary structure of proteins is the homology-based modelling using known structures. A real chance to predict the general fold of a protein arises only in cases with a sufficient sequence homology (e.g. 27% over 100 residues). In this analysis we examine the phenomenon of inverse sequence similarity (ISS) in proteins and its structural meaning. In sequence data bases we found a lot of examples for ISS up to 34% identity over 204 residues and a surprisingly large number of self-inverse protein sequences. By inspection of inverse similar sequence pairs with known tertiary structures we observe that inverse sequence alignments above the threshold indicating structural similarity generally do not imply comparable folds for both. From our analysis we conclude that the straightforward employment of ISS for protein structure prediction fails even above the known threshold for 'safe similarity'.