Figure 5.

Analysis of serum proteins differentially expressed in MCI converters compared to non-converters. (A) Volcano plot of proteins significantly increased in MCI converters compared to non-converters. Turquoise dots represent proteins whose P-value was below the significance threshold of P < 0.05 (n = 205). (B) The optimal number of multivariate proteins to differentiate MCI converters from non-converters was calculated by machines learning. The plot shows AUCs of the training set as features are added to the model, and AUCs of the testing set during cross-validation. The least absolute shrinkage and selection operator feature selection with SVM for prediction were used. The number of proteins identified with this procedure was 15. (C) ROC curve of the test model. The maximum AUC (0.77) was achieved using the following 15 identified proteins: Q9UHD0, Q9UK55, Q9NPH3, Q96PU8, Q8N474, Q8TBE7, Q9NTK1, Q6UWD8, O14548, P43251, P19876.P19875, Q8NBP7, P52907, P00797 and Q9Y5Q6 (UniProt ID). (D–F) The three top scoring networks analysed by IPA using the 205 differentially expressed proteins in MCI converters. Proteins are represented by nodes: upregulated in red and downregulated in turquoise. Additional interacting molecules not included in the SomaScan are marked in white. Each network is displayed as a series of nodes (proteins) and edges (i.e. lines corresponding to biological relationships between nodes). Solid and dotted lines indicate direct and indirect interactions, respectively. (D) ‘Hematological System Development and Function, Organismal Functions, Organismal Injury and Abnormalities’ (score 48; includes 27 focus molecules from the SomaScan panel). (E) ‘Cell Death and Survival, Embryonic Development, Organismal Development’ (score 43; includes 25 focus molecules from the SomaScan panel). (F) ‘Cell-to-Cell Signaling and Interaction, Cellular Function and Maintenance, Inflammatory Response’ (score 21; includes 15 focus molecules from the SomaScan panel).