*4.5. Metabolic Fingerprinting*

MS Data processing was performed with DataAnalysis 4.4 (Bruker, Billerica, MA, USA) using recalibration with sodium formate (Sigma Aldrich), RecalculateLinespectra (threshold 10,000), and FindMolecularFeatures (0.5–25 min, S/N = 0). Bucketing was performed using ProfileAnalysis 2.3

(Bruker, Billerica, MA, USA) (30–1080 s, *m*/*z* 100–1600, Advanced Bucketing with Δ12 s and Δ5 ppm, no transformation, Bucketing basis = H+). The bucket table was subsequently used as input for analysis via R.R (version 3.6.0) [44] with libraries readr [45], coop [46], gplots [47], data.table [48], parallelDist [49], and devtools [50] were used. For heatmap-generation with several sidebars, a variation of heatmap.2 by Griffith [51] was used. The script used in this publication is deposited on GitHub [repository https://github.com/christoph-hartwig-ime-br/cosine-V3; https://dx.doi.org/10.5281/zenodo.4320539]. For sample comparison, the cosine similarities (dot product of vectors) between samples were calculated. Samples were sorted according to clustering results and pairwise similarities were used to determine metabolic groups. If the pairwise similarity between subsequent clustered samples is 0.7 or higher, they were assigned to one metabolic group.
