Biology Reference
In-Depth Information
to those that meet the requirements of the application area. Defining
appropriate constraints and ensuring that the molecules generated are
consistent with the constraints becomes very important. Focusing the
design according to physicochemical constraints can be an effective
way to prevent inappropriate molecule candidates, guide the search in
space, and derive a QSAR that determines the immunological prop-
erties of proteins.
The process begins with the conception of a mathematical model
that relates the chemical features of the molecule (or series of mole-
cules) to a biological activity, such as immunogenicity or antigenicity.
Without any detailed understanding of the process(es) responsible for
the activity, the model is refined by examining structural similarities
and differences for molecules with variations in immunological prop-
erties. Molecules are selected to maximize the presence of functional
groups and structural features believed to be responsible for the
desired activity. Such a strategy can provide the knowledge and com-
putational capability for de novo design and engineering of protein
molecules with desired immunological properties. The following is a
discussion of some of the computational and mathematical methods
used to derive QSAR. These include artificial neural networks, evolu-
tionary algorithms, and Bayesian networks.
Artificial Neural Networks
Artificial neuron model. The artificial neuron model is a fairly new
form of artificial intelligence (AI) based on brain theory. 110-112 An arti-
ficial neural network (ANN) is a mathematical model of neuron oper-
ation that can, in principle, be used to compute any arithmetic or
logical function. It consists of two types of components or elements:
1) processing, or computation, units (also called “neurons”); and
2) connections, with adjustable “weights” or “strengths.” Figure 1(A)
shows the model of a single neuron (the basis of an ANN) and the
two mathematical terms that describe an artificial neuron k . The fun-
damental operation of an artificial neuron is carried out in three steps.
First, a signal x 1 at the input of synapse j connected to neuron k is mul-
tiplied by the synaptic weight w kj . The synaptic weight of an artificial
Search WWH ::




Custom Search