Statistical feature selection from chaos game representation for promoter recognition

Abstract

Recently, the recognition of promoters has attracted many researchers’ attention. Unfortunately, most previous prediction algorithms did not provide high enough sensitivity and specificity. The aim of this dissertation is to provide a distinct classification between Promoter and non-promoter sequences. We do not consider some well-known patterns around TSS, such as TATAAT-box and TTGACA-box, which were previously used by many researchers. The accuracy of promoter prediction is based on two factors, i.e., the representation of the given DNA sequence and the essential features of the sequence. A Chaos Game Representation (CSR) is adopted for transforming a DNA sequence having promoters and non-promoters into an image. The essential features of the CGR are selected by applying the concept of statistical feature selection. It is aimed at finding the Smallest set of features that can distinguish the classes over the full set and reduce the dimension of the classifier. Recognition can then be performed by a supervised neural network. The method in this dissertation can be applied to both prokaryotic and eukaryotic Organisms.