The role of long-range interactions in defining the secondary structure of proteins is overestimated

Secondary structure predictions based on the properties of individual residues, and sometimes on local interactions, usually fail to exceed 65% efficiency. Therefore, non-local, long-range interactions seem to be a significant cause of this limitation.

In this paper, we apply approaches to localize highly interacting residues and clusters of residues involved in multiple non-local interactions, and test various secondary structure predictions on this separate subset to assess the effect of long-range interactions on the prediction efficiencies. It was found that only a marginal part of the failure of secondary structure predictions results from the presence of long-range interactions. Alternative possibilities are also discussed.