MULTISTAGE FILTERS FOR IDENTIFICATION OF EUKARYOTIC PROTEIN CODING REGIONS

Abstract

A class of multistage filters, namely, real narrowband bandpass filter (RNBPF) has been previously used for identification of protein coding regions. This filter passes the frequency component at 2π/3 along with its conjugate. This conjugate frequency component may degrade the identification accuracy. To improve the identification accuracy, two types of multistage filters are proposed in this paper. A complex narrowband bandpass filter (CNBPF) is proposed for suppressing the conjugate frequency component which, in turn, reduces the background noise present in the deoxyribonucleic acid (DNA) spectrum and improves identification accuracy. By cascading RNBPF with moving average filter (RNBPFMA), another type of multistage filter is proposed. As moving average filter smooth out the rapid variations in the DNA spectrum, RNBPFMA improves the identification accuracy. The computational complexity of RNBPFMA is less than that of CNBPF. The RNBPF and proposed multistage filters are compared with previously reported short-time discrete Fourier transform (ST-DFT) method in terms of computational complexity. It is found that multistage filters reduce the computational load to a greater extent compared to ST-DFT method. The identification accuracy of the proposed CNBPF and RNBPFMA methods is compared with existing anti-notch filter and RNBPF methods. The results show that proposed methods outperform existing methods in terms of identification accuracy for benchmark data sets.