Reaction databases provide a rich
source of information on organic reactions. A
combination of a Bayes classifier with a
self-organizing neural network is used to group
individual reactions into reaction types on the basis
of physicochemical description of the reaction center.
This allows to detect and visualize the important
characteristics - and the driving forces - of a class
A set of chemical reactions characterized by
physicochemical properties of the atoms and bonds of
the reacting center is entered into a Kohonen neural
network. This results in a two-dimensional landscape of
organic reactions. Similar reactions are grouped into
reaction types, dissimilar reactions are separated from
each other. Furthermore, this method can recognize
special reactions, thus outlining the scope of a
reaction type and can locate unusual reactions.
The automatic classification of reactions can be used
for an efficient reaction database searching and derive
knowledge on chemical reactions form a series of
individual reactions. Such knowledge can be utilized in
systems for reaction prediction (such as EROS) and synthesis planning (WODCA).
L. Chen, J. Gasteiger
Organic Reactions Classified by Neural Networks:
Michael Additions, Friedel-Crafts Alkylations by
Alkenes, and Related Reactions,
Angew. Chem. Intern. Ed. Engl.,
1996, 35, 763-765; Angew.
Chem., 1996, 108,
L. Chen, J. Gasteiger,
Knowledge Discovery in Reaction Databases: Landscaping
Organic Reactions by a Self-Organizing Neural
J. Am. Chem. Soc., 1997,
H. Satoh, O. Sacher, T. Nakata, L.
Chen, J. Gasteiger, K. Funatsu
Classification of Organic Reactions: Similarity of
Reactions Based on Changes in the Electronic Features
of Oxygen Atoms at the Reaction Sites,
J. Chem. Inf. Comput. Sci.,
1998, 38, 210-219.