Applications of Artificial Neural Networks in Genome Research

Abstract

Applications of artificial neural networks in the field of genome research will be reviewed and some more recent developments in neural network research relevant for future applications will be surveyed. The basic definitions for artificial neural networks and neural learning algorithms will be introduced. The applications range from the recognition of translation initiation sites in nucleic acid sequences, the recognition of splice junctions and exons/ introns in mRNA, the detection of uncommon sequences in cDNA, to the prediction of secondary and tertiary structures of proteins from the amino acid sequence, the detection of structural motifs in protein sequences and the classification of protein sequences into functional families. Most applications employ multilayer feedforward networks trained supervised with the backpropagation learning algorithm or self-organising Kohonen maps adapted unsupervised for feature extraction. The most promising developments in neural network research usable in all mentioned applications are new modular network architectures with more problem-tailored connection topologies such as linked receptive fields and recurrent networks with short-term memory capable of modelling any dynamical system using only inductive learning.