State of the Human Proteome in 2014/2015 As Viewed through PeptideAtlas: Enhancing Accuracy and Coverage through the AtlasProphet

The Human PeptideAtlas is a compendium of the highest quality peptide identifications from over 1000 shotgun mass spectrometry proteomics experiments collected from many different labs, all reanalyzed through a uniform processing pipeline. The latest 2015-03 build contains substantially more input data than past releases, is mapped to a recent version of our merged reference proteome, and uses improved informatics processing and the development of the AtlasProphet to provide the highest quality results. Within the set of ~20,000 neXtProt primary entries, 14,070 (70%) are confidently detected in the latest build, 5% are ambiguous, 9% are redundant, leaving the total percentage of proteins for which there are no mapping detections at just 16% (3166), all derived from over 133 million peptide-spectrum matches identifying more than 1 million distinct peptides using AtlasProphet to characterize and classify the protein matches. Improved handling for detection and presentation of single amino-acid variants (SAAVs) reveals the detection of 5,326 uniquely mapping SAAVs across 2,794 proteins. With such a large amount of data, the control of false positives is a challenge. We present the methodology and results for maintaining rigorous quality, along with a discussion of the implications of the remaining sources of errors in the build. We check our uncertainty estimates against a set of olfactory receptor proteins not expected to be present in the set. We show how the use of synthetic reference spectra can provide confirmatory evidence for claims of detection of proteins with weak evidence.