Mass spectrometry image normalization (source code)

= Mass spectrometry image normalization

\b[Mass spectrometry image normalization] rescales the spectra associated with individual image pixels to make a specified comparison more meaningful. In <mass spectrometry imaging>, raw intensity depends on both the sample and measurement conditions; <MALDI matrix> distribution can change the recorded intensity. A normalization rule therefore embodies an assumption about which variation should be removed.<Normalization in MALDI-TOF imaging datasets of proteins: practical considerations by Sören-Oliver Deininger and others>[Introduction and “A pledge for normalization”: preparation-dependent signal variation and the assumptions of spectrum-wide scaling.]{ref}

For example, <total ion current normalization> divides each nonzero spectrum by the sum of its included intensities. Write pixel $p$ as values $I_{p,k}$ and set
$$
J_{p,k}=\frac{I_{p,k}}{\sum_j I_{p,j}},\qquad \sum_j I_{p,j}>0.
$$
In an invented two-feature example, spectra $(2,6)$ and $(4,12)$ both become $(1/4,3/4)$. This removes a common multiplicative change. But $(2,14)$ becomes $(1/8,7/8)$: the first feature now appears half as strong as in $(2,6)$ even though its raw intensity is unchanged. The normalized image represents a relative share of the included signal. It answers a different question from an image of raw intensity or calibrated molecular abundance.

In actual <MALDI> datasets, Deininger and colleagues found that locally dominant signals could introduce artifacts under total-signal normalization; they compared median and noise-based alternatives. Their examples motivate inspecting the denominator as well as the selected ion image.<Normalization in MALDI-TOF imaging datasets of proteins: practical considerations by Sören-Oliver Deininger and others>[Abstract and Figures 2–4: differing tissue profiles and normalization-induced artifacts.]{ref} Specify the mass range, excluded peaks and zero-signal handling. Neither <total ion current normalization> nor another single scalar correction can, in general, undo selective suppression of one molecular signal.