Biology Reference
In-Depth Information
packages, including ProtScale,
134
which assigns a numerical value to each
type of amino acid as predefined by any of the >50 scales of physico-
chemical properties and conformational parameters. Thus, substituting
the string of amino acid characters with the corresponding listed values
or selecting the window size and running the program to compute the
profile can transform any given peptide or protein sequence into a
numerical/signal vector. Software tools for the automatic modeling of a
protein's three-dimensional structure are also available on the Internet.
These web-based computational resources are designed to provide
atomic coordinates that can be used as a Cartesian three-dimensional
representation of a given sequence, which in turn can be used for ANN
processing. Examples of web-based servers for homology modeling of
three-dimensional structures of proteins include: Geno 3D (http://pbil.
ibcp.fr/htm/index.php), from the Institute of Biology and Chemistry
of Proteins, Lyon, France; and SCRATCH (http://www.igb.uci.edu/
servers/psss.html), from the Institute for Genomics and Bioinformatics,
University of California, Irvine.
Data post-processing and computational requirements.
Post-
processing is the conversion of the ANN output to a different or
usable form. It comprises output encoding, which is much more
straightforward than pre-processing data encoding. For example, if
the task is to classify a signal vector as yes or no, then only one out-
put neuron, mapped to 1 for “yes” and 0 for “no,” may be used.
The computational requirements for ANN applications depend on
factors such as the type of ANN model, magnitude of the data set, and
the learning task. ANN technology has advanced quickly, generating a
market of both proprietary and freeware systems that make ANN fairly
accessible and easy to implement. In most cases, ANN applications
can be carried out on a desktop computer. Shareware and freeware
ANN simulator software options include Neural Network Toolbox
for Matlab and the freely available Stuttgart Neural Network Simula-
tor (SNNS) (http://www-ra.informatik.uni-tuebingen.de/SNNS/).
Additional software can be found at the website http://www.geocities.
com/fastiland/NNwww.html, hosted by Professor Andrés Pérez-Uribe,
University of Applied Sciences of Western Switzerland, Yverdon-les-Bains;
and at http://www-dsi.ing.unifi.it/neural/w3-sites.html#software from
