Optimizing property codes in protein data reveals structural characteristics

We search for assignments of numbers to the amino acids (property codes) that maximize the autocorrelation function signal in given protein sequence data by an iterative method. Our method yields similar results to optimization with the related extended Jacobi method for joint diagonalization and standard optimization tools. In nonhomologous sets representative of all proteins we find optimal property codes that are similar to hydrophobicity but yield much clearer correlations. Another property code related to alpha-helix propensity plays a less prominent role representing a local optimum. We also apply our method to sets of proteins known to have a high content of alpha- or beta-structures and find property codes reflecting the specific correlations in these structures.