Computational methods for thermal stability proteomics

In the last decades, molecular biology has transformed into a data-rich discipline. This trend is driven by developments in imaging and the continuous increase in available omics technologies which allow for high-throughput profiling of various types of molecules in a given biological system. Classical omics approaches profile the abundance of thousands of cellular biomolecules, e.g., RNAs or proteins. Recently developed assays, such as Thermal Proteome Profiling (TPP), however, can additionally inform on biophysical states of proteins. By choosing the right experimental design or through contextualization of TPP experiments they can reveal small molecule protein engagement, protein-protein interaction (PPI) dynamics or effects of post-translational modifications (PTM). However, while experimental de- signs, reproducibility, amenable organisms and throughput of the TPP assay are being advanced at a fast pace, computational methods for statistical analysis of obtained data are lagging behind. This thesis proposes a suite of computational methods to provide tools for several of the aforementioned application areas of TPP. First, it describes a software package for analysis of TPP experiments in the context of PPIs and suggests a method for detection of differential PPIs across conditions. The application of this method to different TPP datasets revealed significantly changing PPIs during different phases of the human cell cycle and behavior of protein complexes in Escherichia coli within and across cellular compartments. Second, this work addresses a specific experimental TPP setup called 2D-TPP in which thermal stability of proteins is measured as a function of temperature and concentration of a compound of interest to find proteome-wide interactions of the compound. This was done by implementation of a curve-based hypothesis test to analyze data obtained from such experiments with false discovery rate control. The method was benchmarked on simulated data and on several real datasets. Application of the software to 2D-TPP datasets profiling epigenetic drugs revealed hitherto unknown off-targets and downstream effects of these drugs. Third, the same computational method was applied to a 2D-TPP dataset profiling ATP and GTP in a crude cell extract. The analysis of these datasets revealed functional roles of ATP in proteome regulation ranging from allosteric binding, over protein complex assembly and condensate formation. Last, a method for analysis of TPP experiments to profile the effect of PTMs is presented. While the application of this method led to the detection of phosphosites known to be involved in protein regulation, it also pointed out sites which appear to be involved in controlling the localization of proteins to membrane-less organelles. Taken together, this thesis introduces and showcases computational methods for different application areas of TPP. The presented methods are implemented as open source software packages to enable long-term availability and access to the broader community.