Histograms to assess protein enrichment.

<p>The proteins were grouped for their abundances and the relative frequency with which different proteins in the abundance bins occurred. <b>A.</b> Bias of detection. In shotgun proteomic experiments not all proteins present can be detected. The subpopulation of identified proteins is composed of more abundant proteins than the average of all proteins present. Likewise, the subpopulation of quantified proteins is biased to the more abundant proteins among the identified proteins. The reason for the prevalence to identify abundant proteins is that abundant proteins are most likely yielding more intense peptide precursors and are hence most likely selected for fragmentation. Likewise, abundant proteins are most likely generating more and higher scoring peptide identifications, explaining the bias (modified from [<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0159824#pone.0159824.ref052" target="_blank">52</a>]). <b>B.</b> Observation of enrichment. Upon enrichment, proteins can be shifted either within the observation window (1) or into another (2). Moreover, proteins that are not promoted by the enrichment procedure can be decreased in intensity and even leave the observation window. <b>C.</b> Enrichment analysis comparing the mean intensity of a subpopulation of identified proteins with the overall mean. This leads to information about the sample composition but as a chemical enrichment procedure does not necessarily lead to such an intrasample shift of the intensity distribution the method’s axiom seems improper. As an example, if an absolute enrichment had been achieved leading to the identification of proteins from a specific gene ontology only, this method would yield no enrichment at all. <b>D.</b> Schematic representation of all 20,256 proteins annotated in UniProt/Sprot and the 22.4% of these, which were assigned as plasma membrane proteins (PM). As there was no prior information about abundances in the data bank, p-value-based enrichment calculations assumed a uniform distribution, which was very unlikely to reflect the real situation. The strategy presented here was to calculate p-values for cell surface proteome enriched samples as well as for total proteome samples and compare p-values of respective intensity bins.