Estimation and use of protein backbone angle probabilities.

A procedure is described for estimating the probabilities for the backbone phi-psi angles of a protein molecule from the data base of known protein structures. The procedure is basically an adaptation of a published secondary structure prediction scheme, applied to the phi-psi angle bins rather than to the secondary types. The phi-psi angle probabilities estimated this way include all effects of local sequence and are "context sensitive" in that the probabilities for a given residue type depend on its position along the sequence. These probabilities can be used to predict the three-dimensional structure of short polypeptides that are stabilized mainly by local interactions only and to predict the protein folding initiation sites, with moderate to good success rates in each case. They are also potentially useful for efficient sampling in a Monte Carlo scheme of protein tertiary structure prediction methods.