Hurst CGR (HCGR) - A novel feature extraction method from chaos game representation of genomes

Abstract

The performance of a classifier depends on the exactness of the feature vectors extracted from the dataset. Here, a novel method for feature extraction from genome sequences is presented which combines Chaos Game Representation (CGR) and Hurst exponent. The former maps genome sequences into fractal images while the latter acts as a quantifier for such images. The suitability of the new feature vector is attested by classifying 8 categories of eukaryotic genomes accessed from NCBI. The classification results prove that application of Hurst exponent over Chaos Game Representation formats of genome sequences can extract signature features representative of the underlying sequences, thus presenting HCGR as a new feature for classification of genomes. © 2011 Springer-Verlag.