Biology Reference
In-Depth Information
mathematical technique known as Bayesian inference have improved
the performance of many AI programs to the point that they are
being used in the real world. For example, the Microsoft Office 97
paper-clip assistant is based in Bayesian networks (marking Microsoft's
first commercial deployment of Bayesian models).
250
The application
of BN methods is steadily on the rise in biomedical research. They have
been used for protein folding and secondary structure predictions,
251-253
protein-protein interactions,
254-256
gene-drug dependencies,
257
identi-
fication of transcription sites
258
and splice sites,
259
estimation of gene
regulatory networks,
260-263
and a range of other areas including med-
ical diagnosis.
264;,265
A Bayesian network (also known as a Bayesian belief network,
probabilistic causal network, chain graph, and knowledge map) is a
form of artificial intelligence designed to cope with uncertainty.
266
Such systems operate using “Bayesian reasoning” under the Bayes rule
principle.
267-270
The BN method has both quantitative and qualitative
attributes. With respect to the latter attribute, a BN is represented as
a directed acyclic graph in which the nodes represent random vari-
ables of interest and the edges (arcs) linking them represent causal
influence. In the simplest form of BN, quantitatively each variable is
considered to be independent of its ancestors given its parent, where
the parent/ancestor relationship is with respect a fixed topological or
hierarchical ordering of the nodes. The BN represents a local, rather
than a global, probability model, in which each node consists of a set
of conditional independent probability distributions (Fig. 6). Among
the advantages of Bayesian networks is their ability to: 1) combine
highly dissimilar types of data (i.e. continuous or discrete data; or data
from disparate sources) into a common probabilistic framework; 2) pro-
vide an intuitive and interactive graphical model representation that
allows the researcher to read true probabilities from the graphical
structure to determine the significance of each feature, which can help
explain the relevance of peptide/protein structural features for a given
problem application; 3) be used to discover data structure and condi-
tional probabilities directly from raw data (
tabula rasa
) through unsu-
pervised learning; and 4) provide a conceptual framework for the
clear distinction between causal and statistical parameters and general
concepts extending from this distinction, thus providing a proper tool
