[Next] [Previous] [Up] [Top]

4.3 Prediction of Secondary Structure

4.3.2 statistical

Statistical based methods of secondary structure prediction were among the first to be developed, following the determination of protein three-dimensional structures. Of the twenty naturally occurring amino acids a number do demonstrate statistically significant conformational preferences. For example, Glu, Ala, Leu, Met, Gln, Lys, and Arg are preferentially found in helices (from 1.6 to 1.2 times more frequent, in decreasing order) whereas Val, Ile, Tyr, Cys, Trp, Phe, and Thr are more frequently found in strands (1.9 to 1.2 times more frequent) than in non-helical segments. The residues Gly, Asn, Pro, Ser, and Asp are 1.8 to 1.2 times more likely to be found in turn segments than any other structure.

A bit of reflection over these preferences will assure the reader that based on the examples we have seen of particular amino acid sidechain interactions in secondary structures, a good number of these preferences make sense. For example, The charged amino acids prefer helical conformations due at least in part to the strong preference of helices to be located at the protein surface ( and and favorable interactions with helix dipoles ( and the hydrophobic residues prefer beta strands which are most often buried in the interior of globular protein.

Perhaps the simplest and most commonly used (and also abused) method of secondary structural prediction is that of Chou & Fasman (1978). The authors found that fairly simple rules could be developed to locate secondary structural elements based on the tendency of segments of the polypeptide chain with similar amino acid preferences to correlate with that secondary structure type in known structures. In short, the Chou and Fasman method involves first locating helix and sheet "nuclei" . Helix nuclei consist of hexapeptides with at least 4 residues of high helix potential. Similarly, strand nuclei contain three residues of high strand potential in a pentapeptide sequence. These nuclei are then extended in the N-and C-terminal directions until the average tetrapeptide falls below a threshold. A number of improvements have been proposed in the years since the original method was proposed. These improvements include (among many others) position dependent conformational preferences. Another commonly used statistical based algorithm is that of Garnier, Osguthorpe & Robson (1978) which includes directional information.

The success of these algorithms is difficult to judge as a number of the proposed algorithms are open to different interpretations and hence different groups using similar methods obtain different results. In their analysis of three commonly used prediction methods, Kabsch & Sander (1983) found that none achieved better than 56% accuracy in a three state model (helix, sheet, other).

No Title - 31 MAY 96
written by Kurt D. Berndt

[Next] [Previous] [Up] [Top]

VSNS Main Index

Generated with CERN WebMaker