Graph kernels in chemoinformatics
Résumé
Graphs provide a generic data structure widely used in chemo and bioin-
formatics to represent complex structures such as chemical compounds or
complex interactions between proteins. However, the high flexibility of this
data structure does not allow to readily combine it with usual machine learn-
ing algorithms based on a vectorial representation of input data.
Graph kernels are defined as similarity measures between graphs. Under
mild conditions, graph kernels correspond to scalar products between possi-
bly implicit graph embeddings into an Hilbert space. Thanks to this graph embedding, machine learning methods which may be rewritten so as to use
only scalar products between input data, such as SVM, can be applied on
graphs. Graph kernels thus provide a natural connection between graph space
and machine learning.
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...