Papers of interest:
In some recent work, Noble and colleagues describe the way they
develop kernel methods that generalize their (and other's) earlier
work on SVMs. Kernel methods allow one to express the similarity
between a pair of objects with respect to various different types of
data (e.g. expression correlations or similarity in hydrophobicity) in
a uniform framework, and then to combine all these methods of
similarity in an optimized fashion. Lanckreit et al describe an
overall formalism for using these kernel methods in the framework of
bioinformatics. Ben-Hur & Noble describe how these kernel methods can
be specifically adapted to protein-protein interactions where one is
computing the similarity between pairs of proteins rather than
individuals. Tsuda & Noble describe a particular method of kernel
construction.
--
Tsuda K, Noble WS. Learning kernels from biological networks by
maximizing entropy. Bioinformatics. 2004 Aug 4;20 Suppl 1:I326-I333.
Ben-Hur A, Noble WS. Kernel methods for predicting protein-protein
interactions. Bioinformatics. 2005 Jun 1;21 Suppl 1:i38-i46.
Lanckriet GR, De Bie T, Cristianini N, Jordan MI, Noble WS. A
statistical framework for genomic data fusion. Bioinformatics. 2004
Nov 1;20(16):2626-35. Epub 2004 May 6.
