UC San Diego

High-throughput tandem mass spectrometry has enabled the detection and identification of over 75\% of all human proteins predicted to result in translated gene products from an available tens of terabytes of public data in thousands of datasets. This thesis explores what we can learn from this, as well as the challenges that arise when considering proteomics data at a repository scale. First, we will consider validating what is known, through resources to build, curate, and explore both FDR-controlled and user submitted libraries. Second, we present a tool that allows for an automation of application of strict community guidelines criteria to any set of search results, including peak quality and novel FDR controls. Third, we introduce a method to illuminate the extent of what is not yet known using a new clustering approach designed to explicitly model peptide diversity by explicitly modeling spectrum coelutions. Finally, fourth, we developed a method for extremely fast single spectrum searches against spectrum repositories consisting of billions of spectra to both confirm or refute knowledge base IDs as well as discover similar spectra to those consistently unidentified.