Gaussian mixture modeling of alpha-helix subclasses: structure and sequence variations.

Abstract

Classification of helical structures and identification of class specific sequence features is of interest for protein structure modeling. We use geometric invariant based method to first select helix-like local conformations. These conformations are mapped in a principal component space and subjected to Gaussian mixture modeling. The largest Gaussian corresponds to the regular alpha-helix. Kinked helix and curved helix appear as a separate gaussians. Class conditional, position specific amino acid propensity analysis reveals striking difference among the three classes. In regular helix, proline propensity is significant only in the beginning and low in the rest of the region regardless of length of the helix. In kinked helix, the proline propensity has a sharp peak at the helix center, while in the curved helix, the proline propensity has a broad peak in the middle region.