In the fourth part of the lecture, we'd like to describe now the progress which was done over the last five years of this project. The first was we established a work flow, a work flow of omics data generation. We established how to handle cell samples to run targeted and untargeted acquisition of metabolomics data, of transcriptomics data. How to analyze these and identify then what are the significant changes and lead them towards a positive identification. I'll come back to some of these aspects in a second. And we also established a work flow, an integrated biology work flow, which allows to use both metabolomics data and microarray data. We should also sample to [INAUDIBLE] and some new next generation sequencing data in order to make sense of this and identify biological pathways. A series of papers was published, this one for example shows how transcritomics leads us to networks. And this led to quite a bit of excitement, archives of toxicology is the leading experimental toxicology journal in our field. And two guest editorials were invited in order to respond to his article, which was showing how to make sense of transcriptomics data and to deduce networks. There's also quite a bit of challenge this prospects as regards to using metabolomics identification. This article, which is currently in press, shows that how different ways of analyzing metabolomics data, can be used or cannot be used for past identification. In short, first of all, the difficulties and limitations of commonly used tools like over representation analysis, ouantitative enrichment analysis and pathway analysis. This is something definitely of interest to the specialists. And the underlying problem is that untarget metabolomics are not restricting ourselves to a few things we want to measure is necessarily producing a lot of noise. We have artifacts, outliers, miss-identified metabolites, and the problem is that already this extracting the metabolites from the cells. This our choice of column for the [INAUDIBLE] in our liquid [INAUDIBLE] systems was the positive or negative [INAUDIBLE] restricting what we can actually measure dramatically. And this is very much biasing what metabolites we can actually identify. And our analysis is based on the [INAUDIBLE] that we showed you earlier. And because of their short coming, we will at best mismatch and very often receive perturbed pathways, which are significant but not relevant for the biology. And I don't want to go into too much detail, but the paper showed for example overall presentation and analysis coming up. In two repeat experiments only was [INAUDIBLE] and very simple generic pathways being perturbed. When using quantitative enrichment analysis, you see very, very little correspondence in the pathway identified in experiment one and two. Very few were common in both and again, they were not telling us really a lot about the estrogenic response of [INAUDIBLE] cells. And the same does hold true for pathway analysis with impala, which is actually combination of four different databases. And you can see despite the fact that we are using such large database, we can to very little produce the password identification and most of these not really informative for [INAUDIBLE]. That is also an upside of we were able to use weighted correlation network analysis and identified 53 metabolized. Which formed a network and this is actually a starting point for hypothesis on the established network. I invite those of you interested in such level of detail to look into the paper to take home messages. It was a very, very painful exercise in demonstrating that a lot of standup methodologies are not really doing a proper job and very well controlled. But at the same time, there is some hope of we demonstrated that technologies, which are namely used for gene expression so far are actually capable of identifying networks. So we believe that using weighted correlation and network analysis approaches that cluster metabolites by network topology. Are more promising than the focus on individuals significantly changed entities metabolize of genes. And these classes can be used for further analysis. You're combining this then with text mining from the scientific literature. Other sources of high throughput data and we can compare this then with other pathway analysis results and map these things against each other. So in consequence, we have produced the type of workflow here. So summarizing where we are at the end of five years of this project, we have shown with multi-omics, karotyping, and competitive genome hybridization. For the first time the full extent of instability of a tumor cell line. And we feel that this is actually only opening up for more problems that you will see in very very commonly used cell lines. We have seen similar publications coming up recently on [INAUDIBLE] cells, cell line, which is used for 75,000 [INAUDIBLE] so far. So it is not a signal phenomenon of MCF 7 cells. Therefore, made progress towards the quality assurance of metabolomics. We have tailored and integrated tools for pathway mapping for in a suite of software tools. We developed some dedicated software tools. Some of them are actually already commercially available by Agilent and we established all together a workflow for [INAUDIBLE] mapping. Which is now with some first examples, a scalable process, which means we can apply this to one pathway after the other, exactly the goal of our project. But a lot is still ongoing, the annotation of pathways and the validation. And I will also talk a little bit in the prospects in the next section about a new EU project, which has just started in January this year with $35 million sponsoring to 39 partners. And which is taking up quite a few of the ideas and concepts of our project. So we learned that the basis for making big sense from big data is good big data. And this holds through for both for the biology and measurements quality assurance is key. Good [INAUDIBLE] practice and good measurement practices. We still believe that untargeted measurements are the starting point to be free of the bias of previous knowledge, but we need a combination. We have to follow up our hypothesis with targeted measurements and for this we need reference data. We need better databases, especially in the area of metabolomics. The concept stance, that we can reduce dimensionality mechanism and identifying biomarkers by doing so, that there is a process of measuring a lot to measuring a few meaningful things. And we believe that the networks and the correlations between the different entities we are measuring. Are more important than the significant changes of individual parameters which we are cherry picking.