Fig. 4: Molecular probabilistic maps (MPMs) are robust against various forms of artificially added noise.
From: Spatial probabilistic mapping of metabolite ensembles in mass spectrometry imaging
a Schematic representation of ion intensity distributions of a typical metabolite-of-interest (MOI) (fMOI(k); dashed black curve) and of the corresponding Gaussian distribution (fGaussian(k); orange) from which artificial Gaussian noise and interference noise (b) were sampled. Mean and standard deviation of fGaussian(k) are equal to those of fMOI (k). fint-artifacts (k) is a uniform rectangular distribution whose range exceeds the range of fMOI (k) (see dotted x-axis). b Gaussian noise is sampled from fGaussian (k) and is added for all pixels to the raw signal of MOI m/z present in each pixel. Interference noise is also sampled from fGaussian(k) but is added for all pixels arbitrarily at MOI m/z + \(2{\sigma }_{G}\) where \({\sigma }_{G}\) is the standard deviation of the Gaussian-weighting envelop. The latter is a function of the mass resolving power at MOI m/z. Intensity artifact noise is sampled from fint-artifacts(k) and is added to n = 10 randomly selected pixels at m/z MOI. c MPMs (middle row) but not ion images (upper row) of a sphingomyelin SM(d34:2)[M+H]+ (m/z 701.5592; FDR ≤ 0.1) are robust against various forms of artificially added noise and signal artifacts: random Gaussian noise (second column), presence of abnormally high-intensity peak artifacts (third column), and added overlapping peaks \(2{\sigma }_{G}\) away from MOI m/z (fourth column). Despite the degraded visual quality of artificially “contaminated” data, MPMs are able to identify areas of significant metabolite spatial relative abundance. This is demonstrated by the high degree of overlap (yellow) for all noise types between estimated MPM hotspot contours of raw (green) and artificially “contaminated” data (red), as judged by their Dice similarity coefficient (DSC).
