GeneScreen is a collection of computational
statistic routines, whose objective is to process gene expression data
(typically from DNA microarray time-course experiments), extracting
significant gene association patterns.
In GeneScreen, the conditional mutual information among genes in a
cluster (also known as the co-information) is used as a measure of
conditional dependency. The mutual information is used as a scoring
metric for its capability of detecting dependencies of high-order
(non-linear), as opposed to a simple correlation measure, which is only
capable of representing linear dependencies in the data.
For a given microarray assay experiment, all possible unique
combinations of three genes are considered and the co-information is
used to assign a score to each such combination. GeneScreen includes a
set of tools devoted at pre-processing of the transcription data,
performing a series of tasks which include pruning the set of genes
according to a user defined criterion (e.g. their sample variance),
correcting for univariate and bivariate outliers, or accounting for
missing values.