Statistical feature selection from chaos game representation for promoter recognition

Abstract

The accuracy of promoter recognition depends upon not only the appropriate representation of the promoter sequence but also the essential features of the sequence. These two important issues are addressed in this paper. Firstly, a promoter sequence is captured in form of a Chaos Game Representation (CGR). Then, based on the concept of Mahalanobis distance, a new statistical feature extraction is introduced to select a set of the most significant pixels from the CGR. The recognition is performed by a supervised neural network. This proposed technique achieved 100% accuracy when it is tested with the E.coli promoter sequences using a leave-one-out method. Our approach also outperforms other techniques. © Springer-Verlag Berlin Heidelberg 2006.