Protein Structure Classification Using Local Holder Exponents Estimated by Wavelet Transform

In this paper we use local Holder exponents to capture local patterns in protein sequences. The numerical sequence of a protein based on a 6-letters model of amino acids is considered as a time series, then its local Holder exponents are estimated using the wavelet transform. The probability density of local Holder exponents is then calculated. The probability density values are then taken as features for a perceptron constructed by neural network toolbox in Matlab to classify proteins from the all-alpha, all-beta, alpha+beta and alpha/beta protein structure classes. Numerical results indicate that all selected large proteins can be classified with 100% accuracies.