"The first phase of this project focused on the question which existing pathway analysis algorithms are applicable to the different types of 'omics data. While ample data exited for microarray and transcriptomics data, evidence for the different types of quantitative proteomics data were still scarce. Therefore, I set up targeted experiments to test how different mathematical models work in the different types of quantitative proteomics data. This led to two publications that improved existing workflows for label-free and label-based quantitative proteomics data.
For label-free quantitative proteomics data, we found that spectrum clustering can greatly increase the quantitation of low-abundant proteins (Griss, Stanek et al., JPR 2019). The workflow created for this approach was integrated in ProteomeDiscoverer, one of the most widely used analysis systems for proteomics data, and is freely available at
http://ms.imp.ac.at/index.php?action=spectra-cluster(se abrirá en una nueva ventana). Additional workflows to perform such analysis using only open-source software were made available as nextflow workflows through our newly created, dedicated github repository
https://github.com/bigbio/nf-workflows(se abrirá en una nueva ventana).
For label-based quantitative proteomics data, we found that no existing workflow existed that contained all relevant steps for the data analysis. This forced many researchers to develop own scripts to use the output of one pipeline in a second one. We therefore created a complete workflow that is able to perform all analyses from the raw peaklist data up to the differential expression analysis of the observed proteins (Griss, Vinterhalter, and Schwämmle, JPR 2019). This workflow is shipped as a docker container to ensure the full reproducibility of the performed analysis. The complete software is again available freely and open-source at
https://protprotocols.github.io(se abrirá en una nueva ventana). The results of these first two projects enabled me to identify pathway analysis algorithms that are suited for different types of 'omics data.
To validate the biological use of these identified pathway analysis algorithms, I performed benchmark experiments on a multi-omics dataset studying the effect of melanoma cells on B cells. This led to the surprising discovery that B cells play a crucial role in the inflammatory tumour microenvironment (Griss et al., Nat Comm 2019). We found a specific B cell subtype that is induced by melanoma cells and responsible to recruit T cells to the tumour. Moreover, this subtype predicts the response of patients to immunotherapy and enhances the activation of T cells through immunotherapy in vitro. Therefore, B cells may be a novel target to improve the efficacy of immunotherapies.
The strong clinical data retrieved from this data greatly supported validity of the chosen pathway analysis algorithms. Based on these results it became clear that the new analysis system should support multiple algorithms and be easily extensible to quickly test newly developed approaches. This led to the development of ""ReactomeGSA"": A web-based pathway analysis system that supports different pathway analysis algorithms, multiple 'omics data types as well as the simultaneous analysis of data from different species. The complete ReactomeGSA system, including the respective R package, code for the analysis system itself, and its web-based implementation, are all available at
https://github.com/reactome(se abrirá en una nueva ventana). The R package is available through bioconductor and therefore visible to thousands of bioinformaticians focusing on the analysis of different types of 'omics data. The web-based analysis service is integrated in Reactome's existing web application which has more than 70,000 users per month."
