This paper models transcription factor binding sites (TFBS) using two-step Z transform. We improve the classical method by considering the dependence of base in different sites in the process of building model. We represent the dependence of base in different sites using two-step Z transform in the process of modeling. First, we transform many TFBS sequences with the same length into a model of 24-demension vector. Second, the TFBS that will be identified are transformed into 24-dimension vector, too. Third, we calculate an angle vector between vector of the TFBS and model vector in different dimension. Finally, we train a back propagation artificial neural net (BP-ANN) using the angle vector to identify the real TFBS. We do experiment with four kinds of TFBS of E. coli 12. The result indicates that the algorithm can effectively improve the sensitivity and specificity of identification.
|