Fig. 2: Assessment of unified feature space by uMAIA’s peak matching and comparison with binning approaches.

a, Illustration of the peak-matching problem to place acquisitions into coherent feature spaces. Ground truth, solutions of the standard binning approach and uMAIA. b, Illustration of sagittal sectioning of zebrafish at 72 hpf for MSI. Numbers indicate section position. c, Example of molecules matched across sections with detected m/z values shown (left) and visualization of the corresponding molecule, m/z = 731.45 (right). The color and number of the section refer to the scheme in b. d, Violin plots reporting the distribution of Jaccard distances between isotopolog M + 0 and M + 1 presence across acquisitions after featurization. Distributions are computed across all pairs of sections for 20 molecules with the highest signal intensity. Horizontal lines indicate means. e, Illustration clarifying the definition of feature ambiguity and score calculation. f, Density plots displaying frequency of different peak-matching outcomes for 50 MALDI matrix ions. y axis quantifies the number of sections from the experiment in b where featurization was successful. Max, maximum. g, Heatmaps of MALDI matrix ions (rows) identified over sections (column) for different binning sizes and uMAIA. Color coded according to whether zero, one, two or three peaks were identified within the bin for a given section. Matrix compounds are displayed as in f. h, Bar plots of the distribution of different peak-matching outcomes. x axis quantifies the number of sections in which a compound was identified across the dataset. Bars are colored according to the average ambiguity within the range. Vertical dashed line indicates total number of sections in the dataset. i, As in h, corresponding to uMAIA-retrieved sets. j, Bar plots of different peak-matching outcomes stratified by features: not detected in all sections (less than 20 sections), all sections (exactly 20 sections) or more than 20 sections for each method.