|
|
Networks are a general way to describe data by trying to reduce the relationships
between the data objects to a sparse network. One of the most general ways to
do this is using Gaussian processes to describe a correlational network. Links
between two objects in a correlational network indicate that the two objects
behave similarly. However, indirect relationships are taken into account, so
links between objects whose similarity can be explained by their simultaneous
similarity to a third objects are not shown.
Such networks are similar to the popular Bayesian networks, but they impose less
constraints and require more computational effort to derive. We have devised
a Monte Carlo approach based on the statistical mechanics of sparse Gaussian
matrices to investigate how ensembles of such Gaussian networks can be calculated
to fit a set of observations.
→Presentations and publications
-
C. Best, J. Apostolakis, R. Zimmer:
A Metropolis-Hastings Algorithm for Inferring
Gaussian Graphical Models of Genetic Networks,
preprint, January 2007.
[Abstract]
-
C.Best
A Monte Carlo approach to Gaussian Graphical Models for Gene Networks
presentation, Los Alamos National Laboratory,
March 21, 2005.
[PDF]
-
C.Best
A Monte Carlo approach to Gaussian Graphical Models for Gene Networks
presentation, BCB-Seminar, MPI for Molecular Genetics, Berlin,
December 1, 2004.
[PDF]
-
C. Best:
Gaussian Graphical Models as Statistical Systems
for Gene Expression Networks
presentation, ECCB/ISMB 2004 - SIG Bioinformatics and Statistical Physics,
July 30, 2004.
[PDF]
|
|
