Based on improved parameters predicting protein fold

Based on the protein sequence, by selecting the amino acid composition, motif frequency, low frequency of power spectral density, the predicted secondary structure and the value of auto-correlation function as characteristic parameters, an approach of support vector machine for predicting 27-class protein folds is proposed. Overall recognition accuracy reaches 65.54% in the independent testing. With the same method the overall accuracy of predicting structure class of 27-class protein folds is 81.46%. Our predictive results are better than pervious results.