Correlations in protein sequences and property codes.

Correlation functions in large sets of non-homologous protein sequences are analysed. Finite size corrections are applied and fluctuations are estimated. As symbol sequences have to be mapped to sequences of numbers to calculate correlation functions, several property codes are tested as such mappings. We found hydrophobicity autocorrelation functions to be strongly oscillating. Another strong signal is the monotonously decaying alpha-helix propensity autocorrelation function. Furthermore, we detected signals corresponding to an alteration of positively and negatively charged residues at a distance of 3-4 amino acids. To look beyond the property codes gained by the methods of physical chemistry, mappings yielding a strong correlation signal are sought for using a Monte Carlo simulation. The mappings leading to strong signals are found to be related to hydrophobicity of alpha-helix propensity. A cluster analysis of the top scoring mappings leads to two novel property codes. These two property codes are gained from sequence data only. They turn out to be similar to known property codes for hydrophobicity or polarity. Copyright 1998 Academic Press Limited.