Promoter prediction using DNA numerical representation and neural network: Case study with three organisms

Promoter recognition in various organisms is an area of interest in bioinformatics. In this paper, a feed-forward neural network classifier is presented to predict promoters in three organisms using a DNA numerical representation approach. The proposed system was found to be able to predict promoters with a sensitivity of 87%, 87%, 99% while reducing false prediction rate for non-promoter sequences with a specificity of 92%, 94%, 99% for the human, Drosophila melanogaster, and Arabidopsis thaliana sequences respectively. The results show that feed-forward neural networks can extract the statistical characteristics of promoters efficiently, and that the 2-bit binary coding for DNA data is suitable for the Berkeley Human and Drosophila datasets and the 4-bit binary is suitable for the TAIR Arabidopsis thaliana data sets. Another result demonstrated here is that the proposed prediction system is reconfigurable and versatile with a reduced architecture and computational complexity.