Outliers are atypical data that don't fit the description of the rest of
the data. They could be atypical signals or measurement errors.
In (cDNA) microarray experiments relative expression levels of thousands of
genes are measured simultaneously. A typical gene has an expression level
within a normal range compared to the control.
Genes whose expressions are extremely higher or
lower than that of the control will be considered outliers. These outliers
are "true signals".
Hybridization noises may also result in very high or low expression
ratio. The outliers due to noise are "false signals" here.
Outliers are frequently removed to avoid skewing the statistics of
the rest of the data.