 Okay. So our next speaker is David Lampater from the Verge Genomics. The title for his talk is a genome-wide association between transcriptor factor expression and the Crompton accessibility reviews Crompton state regulators. Yeah. Hi, everybody. So I'm going to present this work that I did in Switzerland at the University of Lausanne. So basically what it's about. The ENCOLE project sort of showed that DHS regions really nicely delineate where transcription factors bind, right? But what is not quite clear is how is this transition happening from closed chromatin to open chromatin. So what is, for certain transcription factors, it could be that it's cause or consequence whether they bind or not. So one aspect of this that has been studied is pioneer factors. Basically these are transcription factors that bind closed chromatin and then have the ability to open it up to some degree and let other factors bind. And you can ask the same question in the other direction basically, which factors are responsible for stability. So how would you study this with the ENCOLE data that's out there? So the basic idea is that if a factor has the ability to open up chromatin, then increasing the levels of this factor should open up chromatin in the regions where there's transcription factor binding sites for this factor, right? So first what you want to have is a motive accessibility score. So you collect all the motives of a certain transcription factor in the genome. You check the enrichment for this motive in the open chromatin section of the various cell lines. And I can see like in this example like cell line three has heavy enrichment for this motive in the DHS regions, cell line two has very little. And now you compare this to the expression across different genes, right? So you can do an association study for expression across the genes and compare this to the motive accessibility scores. And what you would hope is that if a factor is responsible for the transition between open and closed, for example, if it's a pioneer factor, then the association should be strong and you should see for the annotated transcription factor, it should see a strong signal. So the data that was used is the ENCOLE expression data for 109 cell lines and the corresponding DHS data and the library of motives. And now there's a hiccup basically as you can imagine because these data are collected on the same cell lines and these cell lines are related to each other, you have heavy confounding as you can see. Like if you look at the matrix of correlation matrix of the motive accessibility scores and compared to the expression data, you can appreciate that it has a similar correlation structure. And so you have to control for this. And the strategy that was used was a mixed model approach. So basically what you do is you allow the noise term to have a similar shape as the covariance matrix, the covariance matrix of the expression matrix. And so show how this looks like on an example. So this is transcription factor EBF1. So if you do this kind of regression approach across all the genes and you check the p-value distribution, if you do a standard linear regression, you have heavy inflation and the true factor is somewhere in the middle. And now if you use these mixed model approaches, the inflation goes away. So all the genes follow the null distribution and the correct annotated factor is in the tail. It's not quite significant. If you do additional data munching, you can even get it above the Bonferroni cutoff. Now note here it's because we know the annotated factor. We don't really need above Bonferroni level because we can tell just by it's at the second position in the middle example that something interesting is going on. So now if you do this comprehensively, you can show that you have heavy enrichment for the correct factors. So standard linear regression, you already see enrichment. But then if you do these mixed model approaches, it really goes up. Another question is like, is this biologically relevant? So the one list that you would hope to see enrichment in is the pioneer factors. So there has been a high confidence list published of these pioneer factors and you see very strong enrichment of these factors in the transcription factors that have a motive that have a high score. Of course the ones that have a high score also, there are many more than there are not in this list. If you go into the literature, you see quite a few of them that are sort of handed around in the literature as being potential pioneer factors. Yeah. So I believe that's basically a screening tool. This approach provides interesting candidates for pioneer studies. Yeah, that's all. So thanks to Sven Bergman, Sultan Kutarek, Trico and Daniel. Any questions? All right.