Selection and mapping of DNA structural features for short gene recognition

This paper presents two new approaches for short gene recognition in DNA sequences. Three of fourteen DNA structural features are selected based on their classification power by a modified auto-regressive model method. Experiments on human genome show that the method is superior to existing exon detection algorithms. However, this method requires computing time. To overcome this problem, the DNA structural features are mapped to a set of new values. The three signals generated by the mapped feature values are normalised and averaged before their power spectral density is estimated. The computational complexity is reduced substantially using the feature mapping method. Copyright © 2012 Inderscience Enterprises Ltd.