Fig. 5

Plot of metabolite response to depth and the model classification error distribution. Barplots at the top show the number of MFs (mass features) in each depth response category and are broken down by the classification error types. Compounds were assigned a depth category via Dunn’s post-hoc test for significant differences between the sample depths. Good MF found = true positive, good MF missed = false negative, bad MF found = false positive, and bad MF avoided = true negative according to a 0.5 likelihood threshold. Note that the majority of the features in the “bad MF avoided” category fell into the “All equal” depth class for which there was no significant differences between the depths (1365 MFs) but the x-axis has been truncated at 60 for ease of visualization. The boxplots in the bottom illustrate the depth response type for 4 specific categories, with raw peak area plotted on a log scale against the sample depth (DCM = deep chlorophyll maximum, ~ 110 m, DMS-Ac = dimethylsulfonioacetate). All MFs are shown in the central bottom plot across three axes using the rank-normalized median value at each depth as the coordinate for that axis. Each mass feature corresponds to a point in the plot, and their position on the plot describes the shape of their depth profile. Compounds aligning with the 15m axis correspond to compounds with most of their abundance found in the surface ocean; points far to the right side correspond to compounds that are found only at the DCM; points found at the bottom of the plot are those compounds that increased more or less linearly with depth