Nonlinear methods in the analysis of protein sequences: A case study in rubredoxins

Two computational methods widely used in time series analysis were applied to protein sequences, and their ability to derive structural information no t directly accessible through classical sequence comparisons methods was as sessed. The primary structures of 19 rubredoxins of both mesophilic and the rmophilic bacteria, coded with hydrophobicity values of amino acid residues , were considered as time series and were analyzed by 1) recurrence quantif ication analysis and 2) spectral analysis of the sequence major eigenfuncti ons, The results of the two methods agreed to a large extent and generated a classification consistent with known 3D structural characteristics of the studied proteins. This classification separated in a clearcut manner a the rmophilic protein from mesophilic proteins. The classification of primary s tructures given by the two dynamical methods was demonstrated to be basical ly different from classification stemming from classical sequence homology metrics. Moreover, on a more detailed scale, the method was able to discrim inate between thermophilic and mesophilic proteins from a set of chimeric s equences generated from the mixing of a mesophilic (Rubr Clopa) and a therm ophilic (Rubr Pyrfu) protein. Overall, our results point to a new way of lo oking at protein sequence comparisons.