Predicting the Splice Sites in DNA Sequences Using Neural Network Based on Complementary Encoding Method

The artificial neural network has been applied to predict the splice sites in DNA sequences. To use neural network, DNA sequences must be converted into numerical sequences first. Almost all the existing splice sites prediction methods use the orthonormal encoding technique in which each nucleotide is represented by a 4bit binary string with 3 bits set to 0 and one bit set to 1. But this encoding method increases the computational complexity of the neural network. It also neglects the fact that in DNA structure the nucleotides A and T, G and C are complementary. In this paper we present a complementary encoding method, that's the nucleotide A being represented by 1, T by -1; C by 2, and G by -2. Compared with the traditional encoding method, this complementary encoding method could make much less incorrect non-splice sites assignment and greatly shorten the training time of the neural network.