Pattern recognition in the prediction of protein structure. II. Chain conformation from a probability‐directed search procedure

Abstract

AbstractA procedure that finds the most probable conformational states of a protein chain is described. Single‐residue conformations are represented in terms of four conformational states, α, ϵ, α*, and ϵ*. The conformation of the entire chain is represented by a sequence of single‐residue conformational states; the distinct conformations in this representation are called “chain‐states.” The first article in this series described a procedure that computes tripeptide conformational probabilities from the amino acid sequence using pattern recognition techniques. The procedure described in this article uses the tripeptide probabilities to estimate the probabilities of the chain‐states. The chain‐state probability estimator is a product of conditional and marginal probabilities (obtained from the tripeptide probabilities), with a penalty factor to eliminate conformations containing α‐helices and ϵ‐strands of excessive length. The probability estimator considers short‐range conformational information, medium‐range sequence information and some simple long‐range information (through the restrictions on helix and strand lengths). Energy minimization calculations can be carried out in the region of conformational space corresponding to a particular chain‐state. By selecting the most probable chain‐states, the search can be focused on the most probable, or “important,” regions of the conformational space. These energy calculations are described in the third article of the series. The complete procedure described by the three articles is called PRISM, for pattern recognition‐based importance sampling minimization.