Biology Reference
In-Depth Information
the search for a desired property on a list of peptide candidates that
are more likely to conform to such properties rather than following
the conventional random search (epitope mapping) for natural epi-
topes, which is expensive and time consuming. After generation of the
in silico
library, the property vectors were translated back into amino
acid sequences by selecting the most similar residues at each sequence
position according to their physicochemical properties. The com-
puter-generated variants were synthesized and tested for their ability
to bind to human anti-b
1
-adrenoreceptor antibodies. The
in vitro
assays confirmed the applicability of the mathematical strategy to the
generation of a small focusing peptide library. Several non-natural
peptides were found to have increased antibody-binding ability when
compared with natural epitopes. Thus, the generation and selection of
candidates from a focused library provided an efficient starting point
from which to extract knowledge. This approach also allowed them to
expand the search to a large number of sequences more cost effec-
tively (
in silico
) and in less time (computational process) than would
have been possible with conventional techniques such as random
mapping or DNA shuffling.
(3)
Establishment of QSARs with ANN methods
. Armed with the
information generated from the focused library, a three-layered, feed-
forward network with a sigmoid hidden-unit and a single linear out-
put unit was, under the supervised paradigm, to extract the
underlying QSAR by mapping the sequence vectors (90 peptides and
seed peptide) to their respective semi-quantitative absorbance proper-
ties (defined as high, medium and low binding activities). The trained
network was then used as a fitness function for selection of subse-
quently evolved peptide candidates.
(4)
Evolutionary/mathematical search for new variants with desired
properties
. Once the neural network was trained on the information
generated from the focused library data, it was used as a heuristic fit-
ness function and focused approach for searching in sequence space.
Generally, the expanded sequence space search for the
de novo
design
of peptides starts by querying computer-generated sequences.
Schneider
et al.
(1998(b); 12179-12184) opted for an evolutionary
