Normalization

MS spectra of similar samples are not always quantified within the same amplitude range. A normalization step is needed to compare the real intensities fairly. Using normalization, we convert all the spectra to the same intensity ranges. Many different approaches have been proposed and used to tackle this issue. Of these different approaches, one is emerging as a gold standard: the total ion current (TIC). TIC really encodes the average area under the curve [1].

However, a recent study of eight normalization procedures [10] states that it is better to use a local than a global (such as the TIC) normalization method. In other words, the rescaling parameter of the spectra should be estimated by windowing the $m/z$ axis rather than a global computation over all intensities. In addition, median values have proven to be more robust than averages as scale factors against possible outlying peaks [8], and their use is also very well-established in other normalization processes, like, for example, DNA microarrays [25].

Accepting both observations, we propose combining both approaches in our normalization technique: use local estimators over $m/z$ windows with rescaling to the median value of the TIC. The window width is a free parameter tuned by an expert. As a default value, for MALDI/SELDI spectra we suggest using 200 $m/z$ units.