Optimal precursor ion selection for LC-MS/MS based proteomics

Abstract

Shotgun proteomics with Liquid Chromatography (LC) coupled to Tandem Mass Spectrometry (MS/MS) is a key technology for protein identification and quantitation. Protein identification is done indirectly: detected peptide signals are fragmented byMS/MS and their sequence is reconstructed. Afterwards, the identified peptides are used to infer the proteins present in a sample. The problem of choosing the peptide signals that shall be identified with MS/MS is called precursor ion selection. Most workflows use data- dependent acquisition for precursor ion selection despite known drawbacks like data redundancy, limited reproducibility or a bias towards high-abundance proteins. In this thesis, we formulate optimization problems for different aspects of precursor ion selection to overcome these weaknesses. In the first part of this work we develop inclusion lists aiming at optimal precursor ion selection given different input information. We trace precursor ion selection back to known combinatorial problems and develop linear program (LP) formulations. The first method creates an inclusion list given a set of detected features in an LC-MS map. We show that this setting is an instance of the Knapsack Problem. The corresponding LP can be solved efficiently and yields inclusion lists that schedule more precursors than standard methods when the number of precursors per fraction is limited. Furthermore, we develop a method for inclusion list creation based on a list of proteins of interest. We employ retention time and detectability prediction to infer LC-MS features. Based on peptide detectability, we introduce protein detectabilities that reflect the likelihood of detecting and identifying a protein. By maximizing the sum of protein detectabilities we create an inclusion list of limited size that covers a maximum number of proteins. In the second part of the thesis, we focus on iterative precursor ion selection (IPS) with LC-MALDI MS/MS. Here, after a fixed number of acquired MS/MS spectra their identification results are evaluated and are used for the next round of precursor ion selection. We develop a heuristic which creates a ranked precursor list. The second method, IPS LP, is a combination of the two inclusion list scenarios presented in the first part. Additionally, a protein-based exclusion is part of the objective function. For evaluation, we compared both IPS methods to a static inclusion list (SPS) created before the beginning of MS/MS acquisition. We simulated precursor ion selection on three data sets of different complexity and show that IPS LP can identify the same number of proteins with fewer selected precursors. This improvement is especially pronounced for low abundance proteins. Additionally, we show that IPS LP decreases the bias to high abundance proteins. All presented algorithms were implemented in OpenMS, a software library for mass spectrometry. Finally, we present an online tool for IPS that has direct access to the instrument and controls the measurement.