Figure 3
Analysis of the surface proteome of monocyte subsets. (a) Principal component analysis of all 313 proteins annotated ‘cell surface’, ‘plasma membrane’, ‘extracellular’ or with a short GO term. The grouping of the samples suggested that subset type was the major driver of variability as opposed to donor-specific differences. (b) The number of distinct classes of protein expression between monocyte subsets plotted against within-class variance. Proteins were clustered using a k-means approach with 1–20 classes, and the summed distance of each protein from its cluster centroid was determined. This summed distance becomes smaller as more clusters are added, but the rate of decline decreases with each iteration, eventually reaching a fairly steady rate of decline (orange line) that is reflective of overfitting. Clusters added before this point reflect the underlying structure in the protein data, whereas clusters subsequently added through overfitting add no additional useful information. For these data the point of inflexion was between eight and nine classes, suggesting that there are at least eight distinct surface protein profiles. (c) K-means based hierarchical cluster analysis of the 313 proteins identified at the surface of the different classes of monocyte. Right panels – enlargements of clusters particularly enriched in proteins of one class (the adjacent bar colour indicates where each enlargement matches the original cluster plot). C – Classical monocyte sample, I – Intermediate monocyte sample, NC – Non-classical monocyte sample.
