Technique for predicting an ambiguous nucleotide symbol in a dna sequence

Abstract

DNA sequences obtained from a DNA sequencer usually contain some ambiguous symbol N, which can be interpreted as either A, or T, or C, or G. This ambiguity can effect the informative analysis of the DNA sequence. This research focused on transforming this problem to a problem of recognizing a prefix sequence of symbol N. By our assumption that nucleotides and their positions may be related to their neighboring nucleotides, the relative positions are used as the feature of the sequence during the learning and recognizing processes of a neural network for each nucleotide. However, recognizing these features from a training set may take a lengthy time. The problem of increasing the training speed in forms of parallel recognition was also investigated. Experimenting on four Eschericia coli genomes, we selected similar regions of about 40,000 bases from any regions. Each region can train an artificial neural network to recognize all similarity and predictthe actual symbol of N. From random query testing sets, the recognition accuracy is more than 80%.