Prediction of alpha-helices in proteins with the hydrophobic strip-of-helix template and distributions of other amino acids around the hydrophobic strip.

Upon addition of requirements for highly characteristic distributions of helix-stabilizing residues to an alpha-helix predictor based upon identification of a longitudinal, hydrophobic strip-of-helix pattern, maximal sensitivity and efficiency scores approached 50% levels at a high degree of stringency. The hydrophobic strip-of-helix template ([symbol: see text], joined in a circle) was applied to 247 helices to maximize the strip-of-helix hydrophobicity index (SOHHI; the mean hydrophobicity of residues in [symbol: see text] positions). Statistically significant increases or decreases in the frequencies of certain residues are observed in some [symbol: see text] and [symbol: see text] positions: [symbol: see text] in the helix (including N- and C-terminal [symbol: see text] in the helix): Leu, Ile, Val, Phe, and Met; first [symbol: see text] positions in extensions of the template to surrounding segments: not increased Leu, Ile, Val, Phe, or Met; N-terminal [symbol: see text]: Asp, Glu; C-terminal [symbol: see text] and the first [symbol: see text] after the helix: His, Lys, Arg; N-terminal [symbol: see text]: Asn; C-terminal [symbol: see text]: Gln; smallest residue in longitudinal strip-of-helix: Ala or Val at crossing points between helices. An algorithm was then derived to indicate alpha-helices by merging predictions with templates [symbol: see text] and [symbol: see text], where the SOHHI > or = 3.0 in the Kyte-Doolittle scale. Relative to the baseline prediction using only the template, more elaborate rules using combinations of other structural patterns produced more efficient (49 versus 42%, respectively) but less sensitive (32 versus 42%, respectively) predictions when the prediction was required to overlap substantially with the known helix. These maximal sensitivities and efficiencies imply that some helices may be formed in folding intermediates but are lost upon coalescence of the final form. Better predictions of protein structure from the primary sequence may require modeling of competing interactions of local structures in folding intermediates.