Preview

PDF, English Download (11MB) | Terms of use

Abstract

While many stable protein complexes are known, the dynamic interactome is still underexplored. Experimental techniques such as single-tag affinity purification, aim to close the gap and identify transient interactions, but need better filtering tools to discriminate between true interactors and noise. This thesis develops and contrasts two complementary approaches to the analysis of protein-protein interaction (PPI) networks derived from noisy experiments. The majority of data used for the analysis come from in vitro experiments aggregated from known databases (IntAct, BioGRID, BioPlex), but is also complemented by experiments done by our collaborators from the Ueffing group in the Institute of Ophthalmic Research, Tübingen University (Germany). Chapter 3 presents the statistical approach to the data analysis. It focuses on the case of a single dataset with target and control data in order to determine the significant interactions for the target. The procedure follows an expected trajectory of preprocessing, quality control, statistical testing, correction and discussion of results. The approach is tailored to the specific dataset, experiment design and related assumptions. This is specifically relevant for the missing value imputation where multiple approaches are discussed and a new method, building upon a previous method, is proposed and validated. Chapter 4 presents a different approach for the filtering of experimental results for PPIs. It is a statistic, WeSA (weighted socio-affinity), which improves upon previous methods of scoring PPIs from affinity proteomics data. It uses network analysis techniques to analyse the full PPI network without the need for controls. WeSA is tested on protein-protein networks of varying accuracy, including the curated IntAct dataset, the unfiltered records in BioGRID, and the large BioPlex dataset. The model is also tested against the previous same-goal method. While the function itself proves superior, another major advantage is that it can efficiently combine and compare observations across studies and can therefore be used to aggregate and clean results from incoming experiments in the context of all already available data. In the final part, uses of WeSA beyond wild-type PPI networks are analysed. The framework is proposed as a novel way to effectively compare mechanistic differences between variants of the same protein (e.g. mutant vs wild type). I also explore the use of WeSA to study other biological and non-biological networks such as genome-wide association studies (GWAS) and gene-phenotype associations, with encouraging results. In conclusion, this work presents and compares a variety of mathematical, statistical and computational approaches adapted, combined and/or developed specifically for the task of obtaining a better overview of protein-protein interaction networks. The novel methods performance is validated and, specifically, WeSA, is extensively tested and analysed, including beyond the field of PPI networks.