Fig. 4 Heatmap of 170 most important peaks for the species classification random forest model (peaks with maximum of class-specific mean decrease in accuracy of >0.015 are presented); clustering of species is based on hierarchical clustering (average linkage) of the species-mean Euclidean distance based on the whole peak spectrum, the annotation gives maximum peak intensity of the given m/z peak over the whole dataset; heatmap scaling: 0-0.1 class-specific mean decrease in accuracy, peak intensity scaling: 1-7*10-3 arbitrary unit), species included in this analysis: Acartia bifilosa (Abif), A. clausi (Acla), A. danae (Adan), A. negligens (Aneg), A. longiremis (Alon), A. tonsa (Aton), Calanus finmarchicus (Cfin), C. helgolandicus (Chel), C. glacialis (Cgla), C. hyperboreus (Chyp), Centropages bradyi (Cbra) , C. typicus (Ctyp), C. hamatus (Cham), C. chierchiae (Cchi), Metridia longa (Mlon), M. lucens (Mluc), Pseudocalanus elongatus (Pelo), P. moultoni (Pmou), Temora longicornis (Tlon), T. stylifera (Tsty), Paraeuchaeta norvegica (Pnor), Microcalanus sp. (Mcal ), Anomalocera patersonii (Apat), Nannocalanus minor (Nmin), Eurytemora affinis (Eaff), Limnocalanus macrurus (Lmac), Corycaeus anglicus (Cang)