Cascaded Multi-level Promoter Recognition of E. coli Using Dinucleotide Features

Abstract

Promoter recognition has been attempted using different paradigms such as motif/binding regions alone or whole promoter itself. In an earlier paper, a scheme is proposed to use 2-gram features to represent a promoter. These 2-grams gave a comparable performance with the existing methods in the literature. An in-depth analysis of data sets using 2-grams is performed. The analysis presented a scenario where there is a confusion between a majority of promoters with a minor set of non-promoter and vice versa. In an effort to build a complete classification system, using the majority and minority sets in promoters as well as non-promoters, a multi-level cascading system and Ada-Boost classifier are applied. The results indicate that much further improvement is not possible with the modifications proposed.