DNA numerical representation and neural network based human promoter prediction system

In spite of the recent development of computational methods for human promoter prediction, the prediction performance still needs improvement. In particular, the high false positive rate of the traditional approaches decreases the prediction reliability and leads to erroneous results in gene annotation. To improve the prediction accuracy and reliability, a DNA numerical representation and neural network based approach is studied for characterizing DNA alphabets in different regions of a DNA sequence. Three mapping functions are used for converting the DNA alphabets to numerical values so that discriminative biological features are extracted for promoter prediction. Simulations of the proposed system were carried out using a set of genomic sequences from the human chromosome 22 and it was found to achieve high sensitivity and specificity.