Title: Gaussian Mixture Density Estimation
1Gaussian Mixture Density Estimation Applied to
Microarray Data
Can we derive a probabilistic framework giving
each gene gi a probability pi to be expressed or
not in the given experiment ?
Given an expression profile dataset based on a
microarray experiment
Frequency
Intensity
Problems Saturation effects, Different shapes,
Apparently no overall single parameterized
distribution Solution Gaussian Mixture Model,
fitted by EM algorithm optimizing number of
parameters via Bayesian Information Criterion
(BIC)
2Mixture Parameter
Expectation
Number of Cluster
Variance
Gaussian Mixture Model
Mixture model estimation EM algorithm
Number of Clusters is estimated by Bayesian
Information Criterion 2 log-Likelihood log(N)
parameters
3Highly expressed housekeeping genes
Lowly expressed housekeeping genes
Frequency
Expression profile of bone marrow stromal cells
on cDNA array
Empty spots
Saturation Effect
Frequency
Human Affymetrix chip with spike in controls
Frequency
4(No Transcript)