Investigation of biological processes using selective
chemical interventions is generally applied in biomedical
research and drug discovery. Many studies
of this kind make use of gene expression experiments
to explore cellular responses to chemical
interventions. Recently, some research groups constructed
libraries of chemical related expression
profiles, and introduced similarity comparison into
chemical induced transcriptome analysis. Resembling
sequence similarity alignment, expression
pattern comparison among chemical intervention
related expression profiles provides a new way for
chemical function prediction and chemical–gene
relation investigation. However, existing methods
place more emphasis on comparing profile patterns
globally, which ignore noises and marginal effects.
At the same time, though the whole information of
expression profiles has been used, it is difficult to
uncover the underlying mechanisms that lead to the
functional similarity between two molecules. Here a
new approach is presented to perform biological
effects similarity comparison within small biologically
meaningful gene categories. Regarding gene
categories as units, a reduced similarity matrix is
generated for measuring the biological distances
between query and profiles in library and pointing
out in which modules do chemical pairs resemble.

Through the modularization of expression patterns,
this method reduces experimental noises and
marginal effects and directly correlates chemical
molecules with gene function modules.