Figure 1

(A) The logistic correlation with the target (diagnosis) obtained by neural network implementation on 421 lipid signals, highlighting 96 lipid signals (inputs). The logistic correlation is reported as a numeric value between 0 and 1 that expresses the strength of the relationship between a single input and output variables. (B) Partial Last Square Discriminant Analysis (PLS-DA) scores scatter plot calculated on three components by using the 421 variables. The supervised multivariate analysis shows R2Y = 0.98, while Q2(cum) = 0.463. The PLS-DA shows unambiguous separation between the two clinical groups (black dots for OND patients and red dots for MuS). (C) Volcano Plot built on 421 Variables. The volcano plot is a combination of fold change (FC) and t-tests. The x-axis is log2 FC and the Y-axis is log10 (p-value) obtained by non-parametric test and by assuming unequal variance for both classes. The light blue dots are the non-significant variables and the red dots are significant with the −log10 (p-value) > 1. (D) Venn diagram based on the 96 lipid signals obtained by Neural Network (N.N), 131 signals obtained by PLS-DA and 36 variables obtained by Volcano plot (V.P.). (E) Histograms reporting the relative intensity of the most significant lipid species in MuS and OND CSF. *Means p < 0.05, **means p < 0.001 and ***means p < 0.0001 obtained by Mann-Whitney test. Sphingomyelins and Phosphatidylcholines are reported according their Systematic Names. (F) Combine ROC curve of the selected lipids for the evaluation of predictive accuracy of the model, using 100 permutations tests. The plot shows the AUC of all permutation, highlighting the actual observed AUC (0.942) and the empirical p value (0.02). (G) Plot of the predicted class probabilities of each sample through the 100-cross validation, underlying good predictivity of the proposed model in discriminating MuS patients from OND. Details of the reclassification are reported in the confusion matrix associated (0 means OND and 1 means MuS).