Gibbs sampling procedure
Missing data DR
Model parameters Model structure Cm
genes
…
experiments
Biclustering genes / experiments?
Transpose the matrix
Discretize the data6
Model with multinomial likelihood and Dirichlet prior6 Model with Gaussian
likelihood and Wishart prior5
Microarray data matrix
Sample C1 from its full conditional distribution
Sample Cm from its full conditional distribution
…
Sample 1 from its full conditional distribution5,6
Sample v from its full conditional distribution5,6
…
Sample DR[1] from its full conditional distribution
Sample DR[n] from its full conditional distribution
Collect the samples
Defined number of iterations reached?
C1 = … = Cm = 0?
N
DR[1] = … = DR[n] = 0?
N
Initialization
Maximum number of initializations
reached?
N
Y
Y
Y
Mask the obtained bicluster, and set the number of initializations to 0.
Final determination of the bicluster:
Monte Carlo integration – PME estimates for the posterior
distribution of Cm, , and DR
Y End
1 2
3