Omponents (ICs) corresponding to geochemical materials or processes statistically independent of each other.Data structure of bulk sediment geochemistry. Scatter plots of the major element oxide and REY contents of all samples are shown in Fig. 2 and in Supplementary Fig. S2. Most of the plots exhibit data distributions that are sharply cut off at the edges, often with minor elongated structures, which are far from the theoretical distribution of data that constitute a multivariate buy AZD3759 Gaussian distribution9. Indeed, histograms of the elemental contents indicate multimodal, concave, or long-tailed distributions (Supplementary Fig. S1), reflecting the inherent non-Gaussian structures. These observations justify the application of ICA for evaluating the origin of the compositional variations in various deep-sea sediment samples. Independent Component Analysis. ICA was performed by using 11 variables, including the contents of SiO2, TiO2, Al2O3, Fe2O3, MnO, MgO, CaO, K2O, P2O5, total REY except Ce (REY), and Ce. Cerium is handled separately because its geochemical behaviour differs from that of other REY elements owing to its order CEP-37440 specific redox sensitivity16,17 (Supplementary Fig. S3). The ICA result (Fig. 2, Supplementary Figs S4 6) indicates that the 11-dimensional chemical compositions of bulk sediment samples can be successfully expressed by 7 ICs that collectively account for 97.9 of the total sample variance. It should be noted that the ICs are expressed as a set of new base vectors in the compositional spaces along which the chemical compositions of the samples shiftScientific RepoRts | 6:29603 | DOI: 10.1038/srepResultswww.nature.com/scientificreports/Figure 2. Compositional data of deep-sea sediments and extracted geochemical independent components (ICs). The ICs are projected in representative compositional subspaces of (a) SiO2 aO gO and (b) Fe2O3?Ce EY (except Ce) as vectors showing specific directions along which the original chemical compositions changed. The label of each IC is placed along its positive axis. (c ) show the data projected into IC subspaces. Dotted arrows indicate characteristic compositional changes along each IC axis. Sediment samples with specific compositions are colour coded as shown in the legend. No P-rich (P2O5 > 3 wt. ) sample was found in the Indian Ocean in the present work. Reference materials were not included in the dataset analysed by Independent Component Analysis (ICA) but were projected into IC spaces by using the same linear transformation as that for the sediment data, as shown in Supplementary Fig. S4. The data sources of the reference materials are compiled in Supplementary Fig. S2.independently rather than as geometric points with specific compositions, often referred to as end-members12,14. The ICs cannot be ranked simply according to their proportion of data variance as in the case of PCA because the ICs are independent. Thus, the numbering of ICs is commutative, and there is no way to measure the relative importance between ICs12. The effects of the uncertainties in the chemical analysis on the ICA results were assessed by random data perturbation analysis11,14 (Supplementary Fig. S7). With the exception of IC2 showing sensitive behaviour to the data uncertainty, it was confirmed that the remaining six ICs converged well, and were thus considered to be robust signals.Characterization of geochemical independent components. Interpretation of the result of our geochemical ICA was conducted on t.Omponents (ICs) corresponding to geochemical materials or processes statistically independent of each other.Data structure of bulk sediment geochemistry. Scatter plots of the major element oxide and REY contents of all samples are shown in Fig. 2 and in Supplementary Fig. S2. Most of the plots exhibit data distributions that are sharply cut off at the edges, often with minor elongated structures, which are far from the theoretical distribution of data that constitute a multivariate Gaussian distribution9. Indeed, histograms of the elemental contents indicate multimodal, concave, or long-tailed distributions (Supplementary Fig. S1), reflecting the inherent non-Gaussian structures. These observations justify the application of ICA for evaluating the origin of the compositional variations in various deep-sea sediment samples. Independent Component Analysis. ICA was performed by using 11 variables, including the contents of SiO2, TiO2, Al2O3, Fe2O3, MnO, MgO, CaO, K2O, P2O5, total REY except Ce (REY), and Ce. Cerium is handled separately because its geochemical behaviour differs from that of other REY elements owing to its specific redox sensitivity16,17 (Supplementary Fig. S3). The ICA result (Fig. 2, Supplementary Figs S4 6) indicates that the 11-dimensional chemical compositions of bulk sediment samples can be successfully expressed by 7 ICs that collectively account for 97.9 of the total sample variance. It should be noted that the ICs are expressed as a set of new base vectors in the compositional spaces along which the chemical compositions of the samples shiftScientific RepoRts | 6:29603 | DOI: 10.1038/srepResultswww.nature.com/scientificreports/Figure 2. Compositional data of deep-sea sediments and extracted geochemical independent components (ICs). The ICs are projected in representative compositional subspaces of (a) SiO2 aO gO and (b) Fe2O3?Ce EY (except Ce) as vectors showing specific directions along which the original chemical compositions changed. The label of each IC is placed along its positive axis. (c ) show the data projected into IC subspaces. Dotted arrows indicate characteristic compositional changes along each IC axis. Sediment samples with specific compositions are colour coded as shown in the legend. No P-rich (P2O5 > 3 wt. ) sample was found in the Indian Ocean in the present work. Reference materials were not included in the dataset analysed by Independent Component Analysis (ICA) but were projected into IC spaces by using the same linear transformation as that for the sediment data, as shown in Supplementary Fig. S4. The data sources of the reference materials are compiled in Supplementary Fig. S2.independently rather than as geometric points with specific compositions, often referred to as end-members12,14. The ICs cannot be ranked simply according to their proportion of data variance as in the case of PCA because the ICs are independent. Thus, the numbering of ICs is commutative, and there is no way to measure the relative importance between ICs12. The effects of the uncertainties in the chemical analysis on the ICA results were assessed by random data perturbation analysis11,14 (Supplementary Fig. S7). With the exception of IC2 showing sensitive behaviour to the data uncertainty, it was confirmed that the remaining six ICs converged well, and were thus considered to be robust signals.Characterization of geochemical independent components. Interpretation of the result of our geochemical ICA was conducted on t.