Classification of protein targeting sequences by
self-organizing feature maps
Schneider, G.
F.Hoffmann-La Roche,
Pharmaceuticals Division,
Molecular Design & Bioinformatics,
CH-4070 Basel, Switzerland
Much information about the targeting pathway of proteins is
contained in the N-terminal part of protein precursor sequences.
N-terminal targeting signals directing a newly synthesized
protein towards organelles or the secretory route were analyzed
to further understand and visualize these characteristic sequence
features. Self-organizing neural networks were developed for this
purpose, which were able to perform a non-linear projection of
the multi-dimensional sequence space onto a two-dimensional map.
The maps generated show a clear separation of the different types
of targeting sequences and allow for an interpretation of the
features extracted by the network. It turns out that
physicochemical properties of the N-terminal fragments of protein
precursor sequences contribute to the targeting signal. This
finding is a consequence of a new way of sequence representation
in terms of correlations of residue properties. This technique is
presented, and both specific advantages and limitations of
pattern recognition by self-organizing networks are discussed.
Schneider, G. (1997) Concepts in molecular bioinformatics. BIF
Futura 12, 87-97.
Schneider, G., Sjoling, S., Wallin, E., Wrede, P., Glaser, E.,
von Heijne, G. (1998) Feature extraction from mitochondrial
signal peptidase cleavage sites. Proteins: Struct. Funct.
Genet. (January 1998 issue), in press.
| LOCATION |
DATE |
TIME |
| Lecture Hall II |
Sunday, April 5 |
06:20 pm |