Computational solutions for quality control of mass spectrometry-based proteomics

Mass spectrometry is an advanced analytical technique that can be used to identify and quantify the protein content of complex biological samples. Unfortunately mass spectrometry-based proteomics experiments can be subject to a large variability, which forms an obstacle to obtaining accurate and reproducible results. Therefore, to inspire confidence in the generated results a comprehensive and systematic approach to quality control is an essential requirement.

In this dissertation we present several computational solutions for quality control of mass spectrometry-based proteomics. In order to successfully employ comprehensive quality control procedures to assess the validity of
the experimental results three basic requirements need to be fulfilled: (i) descriptive quality control metrics that characterize the experimental performance should be defined; (ii) the basic technical infrastructure to unambiguously store and communicate quality control data has to be available; (iii) advanced analysis techniques are needed to derive actionable insights from the quality control data.

First, we show how secondary metrics that are not related to the spectral data, such as instrument metrics and environment variables, provide a complementary view on the experimental quality. We present the user-friendly Instrument MONitoring DataBase (iMonDB) toolset to manage and visualize these secondary metrics. Second, we introduce the Human Proteome Organization (HUPO) – Proteomics Standards Initiative (PSI) Quality Control
working group, whose aim it is to provide a unifying framework for quality control data. We show how the standard qcML file format for mass spectrometry quality control data can be used as the focal point of a strong community-driven ecosystem of quality control tools and methodologies. Third, we present an unsupervised outlier detection workflow to automatically discriminate low-quality mass spectrometry experiments from high-quality
mass spectrometry experiments. We show how this workflow can replicate expert knowledge in a data-driven fashion, enabling the substitution of time-consuming manual analyses by automated decision-making. Finally, we show how approximate nearest neighbor indexing can be used to speed up spectral library open modification searching by several orders of magnitude, leading to a record number of spectrum identifications in a minimal processing time.

We conclude with an overview of potential future steps that can be taken to further improve computational quality control methods for mass spectrometry-based proteomics, as well as discussing some of the opportunities to apply advanced machine learning techniques in this field with related challenges.