Protein structural class prediction by combining conditional probability with information content

Protein structural class is the most basic and important classification of protein structures. The prediction technique of protein structural class has been developing for decades. Two popular methods, the amino acid frequency based methods and amino acid arrangement and long-term correlation based methods, were proposed in previous work. However, there are opposite advantages and disadvantages in these methods. The former concentrated on statistical analysis while the latter emphasized long-term and biologically significant research. In this paper, we proposed a new method combining probability and information theory with long-term correlation consideration. The results showed that advantages of both original methods were included in our method and the accuracy is much higher than in the traditional procedures.