Skip to main content
Fig. 2 | Biology Direct

Fig. 2

From: Quantitative proteomics signature profiling based on network contextualization

Fig. 2

Benchmarks on control data comprising technical and biological replicates. a Similarity analysis of controls across all proteins without network contextualization. Green corresponds to similarity while red corresponds to dissimilarity. Samples are clustered using hierarchical clustering using Euclidean distance and Ward’s linkage. b Stability analysis of controls at different cut-offs. Y-axis: Jaccard similarity. X-axis: categorical variables for top n% alphas and top n% top and bottom alphas. The top alphas were very stable, ranging between 85 to 88 % agreement rates. On the other hand, the inclusion of the bottom alphas creates increased instability. This led to a significant drop in agreement rates (70–80 %). The lack of instability at the bottom alphas suggested that although under-expressed proteins might be interesting, noise levels are also very high. c Coefficient of variation versus average abundance for each protein. Lower-abundance proteins exhibited higher instability. This further supports the results from panel D, that bottom alphas should not be used in qPSP. d Similarity maps for biological and technical replicates (based on Pearson correlation of the qPSP’s of these replicates). The controls were largely similar to each other. Rep 1 (Inj 1,2 and 3) was slightly different from the other digestions. But on the whole, similarities ranged between 97 to 99.5 %. e False-positive analysis controls. Controls were randomly assigned into two groups of equal size, and qPSP analysis was performed 10000 rounds. For each round, the number of significant clusters (5 % FDR) was determined. The histograms showed that the number of false positive was well within the expectation levels (Sig Clusters Abs E-level: 19, Observed level: 16; Ratio Sig Clusters E-level: 0.05, Observed level: 0.04)

Back to article page