Improving peptide detection in mass spectrometry-based proteomics

Thesis (Ph.D.)--University of Washington, 2022Over the last 30 years, the field of computational mass spectrometry-based proteomics has made great strides. Specifically, the development of database search engines has allowed for the automatic annotation of observed spectra. In addition, the application of target-decoy competition for the purposes of estimating the false discovery rate of a set of peptide-spectrum matches has been instrumental for improving the statistical evidence for a set of confidently detected peptides. While great advances have been made, additional progress is still possible. This work describes three methods for improving computational proteomics methods. The first method describes a new database score function, combined p-value, that aims to take advantage of two advances in database searching: high-resolution MS/MS spectra and statistical calibration. The next method presents a variant of the target-decoy competition process for estimating the false discovery rate. Specifically, this variant is applicable when a subset of peptides in a sample are relevant to the hypothesis being asked. Finally, the last method describes MS1Connect, which measures the similarity of a pair of proteomics runs for the goal of inferring metadata of proteomics runs. Metadata is information about data. For example, given some data, metadata would include information regarding who generated the data and how the data was generated. Metadata is critical for the proper analysis of proteomics data but often it is missing or incorrect. Therefore, methods are needed that can predict metadata of proteomics data. As part of this method, we have also developed MS1Connect, a new score for measuring the similarity of a pair of mass spectrometry runs. We demonstrate that this score can be used for accurate metadata inference of species labels for mass spectrometry runs