Biology Reference
In-Depth Information
ANN weights allowed identification of the major factors influencing
residue exposure. Such analysis made apparent that hydrophobic
residues two or three amino acids up or down from the central amino
acid residue hinder its surface exposure. In general, they found that
high weights for neighboring hydrophobic residues correlated with
their trend to favor burial of the central residue, while high weights
for neighboring hydrophilic residues correlated with their trend to
favor exposure of the central residue.
Design of Biomolecules with Desired Antigenic
and Immunogenic Properties with Artificial
Neural Networks
ANNs have been successfully applied to a wide range of disciplines,
including aerospace,
141,142
robotics,
143,144
medicine,
145,146
and finan-
cial
147
and weather
148
forecasting. This technique, however, has only
recently been used in molecular biology for quantitative structure-
activity relationships (QSAR) determination.
149-154
A major focus of
research aimed at predicting protein QSAR from sequence alone is
based purely on empirical approaches. That is, most of the research is
based on the use of databases of proteins with known structures and
functions. The aim is to find relevant features in these databases,
which can then be extracted and used to derive rules. The ANN can
use these rules to generalize QSAR models for proteins of unknown
functions. Such an approach remains one of the most practical start-
ing points to gain the knowledge needed for protein design. The
following sections describe some examples of how researchers have
applied the ANN technique for the design of molecules with intended
immunological or antigenic properties, particularly, the design of
T-cell epitopes. Also, computer-based strategies for peptide/protein
and “artificial” protein design will be discussed.
T-cell Epitope Prediction
T cells of the immune system continually check for the presence of for-
eign antigens that may indicate the presence of invading microorganisms
