On the nature of the protein folding code

This paper investigates quantitatively the characteristics of the local folding code. The overlapping four-residue fragments which make up the amino acid sequences of 114 proteins are divided into classes on the basis of the physical properties of their constituent amino acids. The distribution of structural types associated with each class of sequence fragment is determined and compared with an ensemble of random structural distributions of the same size selected from the actual protein structures. A criterion is proposed, based on the relative entropies of the two types of distribution, and on a hypothesis as to the characteristics of fragments which code for local structure, that makes it possible to identify those four-residue sequence elements which encode specific time-averaged structure. It is determined that, by this criterion, only 60-70% of the four-residue fragments encode specific structures. It is suggested that the remaining sequence fragments intrinsically encode susceptibility to conformational alteration under the influence of long-range interactions and that this susceptibility is required for correct folding of the molecule. This feature introduces an inherent indeterminacy into the local folding code. The implications of this observation for the prediction of protein structure by various methods are briefly discussed.