Standard Article

Neural Networks

  1. Cathy H Wu1,
  2. Jerry W McLarty2,
  3. Li Liao1

Published Online: 15 SEP 2010

DOI: 10.1002/9780470015902.a0005268.pub2

eLS

eLS

How to Cite

Wu, C. H., McLarty, J. W. and Liao, L. 2010. Neural Networks. eLS. .

Author Information

  1. 1

    Computer and Information Sciences, Newark, Delaware, USA

  2. 2

    Louisiana State University Health Sciences Center, Shreveport, Louisiana, USA

Publication History

  1. Published Online: 15 SEP 2010

Abstract

Artificial neural networks are a form of machine learning, inspired by biological neural networks, which can be trained by examples to recognise patterns, classify data and make predictions. They have been used in a wide variety of problems in the fields of life sciences, such as predicting protein secondary structures, classifying protein families and identifying single nucleotide polymorphism. Problems in life sciences in general and in bioinformatics in particular present challenges to the design and training of neural networks and to the interpretation of networks outputs as well, including complex structured data and limited amount of training examples. In addressing these issues and limitations, significant progress has been made, including ways to design the neural networks to incorporate prior knowledge, to make automatic formulation of network architecture for optimal performance and to construct hybrid systems to leverage advantages of various learning methods.

Key Concepts:

  • Artificial neural networks are a means of mapping input data to output data in such a manner as to perform complex functions such as pattern recognition, classification, prediction and knowledge extraction, that are increasingly useful in bioinformatics applications.

  • There are numerous kinds of neural networks, classified by their architecture (number of layers, number of processing units, interconnections, etc.) and how they are trained to perform.

  • Neural networks have been successfully applied to numerous areas of genomic and proteomic research; the number of such applications has grown explosively in the past few years.

  • One of the most difficult challenges in constructing and training neural networks is often in determining the content and features of data to be used for training and ultimately using the neural networks, that is feature extraction, preprocessing.

  • Hybrid systems, which combine neural networks with other artificial intelligence and statistical techniques, are a promising area of research.

Keywords:

  • artificial intelligence;
  • classification;
  • encoding;
  • sequence analysis;
  • gene recognition