Computing the geometry of a molecule in dihedral angle space using n.m.r.-derived constraints. A new algorithm based on optimal filtering.

We have developed a method based on optimal filtering to determine the three-dimensional structure of a protein from n.m.r.-derived constraints, using the dihedral angle internal representation of the molecule. It differs from currently proposed methods in that it directly produces estimates of errors on the parameters that are refined, hence providing an image of the minimum that has been found. A similar algorithm had already been proposed using cartesian co-ordinates as independent parameters, encoded in PROTEAN2. We found that using dihedral angles significantly reduces the computational burden of the technique, and provides better control over a priori informations that can be used, such as geometric restrictions for proline residues and informations from vicinal coupling constants. Performance of the method, encoded in FILMAN, is demonstrated by application to the folding of a ten-residue alanine polypeptide, to the geometric cyclization of an 11-residue peptide, as well as on the folding of a medium size protein, i.e. tendamistat. The validity of the error estimates on the dihedral angles produced by FILMAN is discussed.