Prediction of protein coding regions in DNA sequences using Fourier spectral characteristics

Existing discrete Fourier transform (DFT)-based algorithms for identifying protein coding regions in DNA sequences (S. Tiwari et al., 1997, D. Anastassiou, 2001, D. Kotlar et al., 2003) exploit the empirical observation that the spectrum of protein coding regions of length N nucleotides has a peak at frequency k=N/3. In this paper, we prove the aforementioned and several other empirical observations attributed to DNA sequences. Our analytical results lead to faster and more accurate DFT-based algorithms for predicting coding regions.