Application of neural networks, PCA and feature extraction for prediction of nucleotide sequences by using genomic signals

Converting symbolic sequences into complex genomic signals reveals surprising regularities of genomes, both locally and at a global scale. This approach allows using signal processing methods for the handling and analysis of nucleotide sequences, specifically for the prediction of nucleotides when knowing the preceding ones in the sequence. In this paper we propose both Feature Extraction (FE) and Principal Component Analysis (PCA) as methods to efficiently extract the major features of a genomic signal, using then a neural network to predict the next sample in the sequence.