Digital signal processing techniques for gene finding in eukaryotes

In this paper, we investigate the effects of window shape and length on a DFT-based method for gene and exon prediction in eukaryotes. We then propose a new gene finding method which combines the selected time-domain and frequency-domain methods, by employing the most effective DNA symbolic-to-numeric representation examined to date in conjunction with suitable window shape and length parameters and a signal boosting technique. It is shown herein that the new method outperforms major existing approaches. By comparison with the existing methods, the proposed method reveals relative improvements of 15.1% to 55.9% over different methods in terms of prediction accuracy of exonic nucleotides at a 5% false positive rate using the GENSCAN test set.