I am not quite sure what data you are working with. Are you talking about a heatmap available
gitools.org/datasets? Also, if are you adding a header by aggregating values from a heatmap?
Anyhow, the TCGA data heatmaps at
gitools.org/datsets contain expression values that are normalized and centered so that 0 means normal expression, positive values is high expression and negative values are low expression. The high and low are relative to the other samples of the cohort. Adding a
mean expression to rows or columns does not change the scale or unit.