 Welcome to this deeper dive into interaction experiments. Today, data from experiments on protein interactions are easily accessible, both from primary repositories, such as the INDAC database, and from secondary integrative databases, such as STREAM. If you're not already familiar with the STREAM database, I highly recommend that you go watch my brief introduction to the resource before continuing. In this presentation, I'll focus on different types of interactions, the types of experiments that they originate from, and finally, a bit on how to combine them. Starting with interaction types. What is an interaction? In STREAM, you have functional associations. You will also often encounter co-expression networks, but here I'll focus strictly on physical protein interactions. There are two definitions of what a physical protein interaction is. There's the strict definition, which is actual physical contact between two proteins, and there's the looser definition, namely the two proteins are part of the same complex. If you look at the 3D structure of a complex, the difference between the two is obvious. Not all subunits of this complex are in direct contact with each other. On top of this, you have to worry about the dynamics of interactions. You have stable complexes like the one I just showed, but you also have transient interactions between, for example, kinases in their substrates, and you have proteins that come in brief contact with each other, known as co-localization. This brings me to the topic of interaction essays. How do we in practice measure interactions? Ideally, we would always determine the 3D structures of the complexes. However, that's not possible because it's too low throughput. For this reason, we rely very much on high throughput essays instead. I'll start with the so-called binary essays, the most famous of which is the yeast to hybrid essay. The idea is that you have two proteins that you want to test the interaction of, the bait and the prey. You fuse the bait protein to a DNA binding domain and the prey protein to a transcription activating domain. If the bait and the prey bind to each other, you will reconstitute a transcription factor that can activate the transcription of a reporter gene. That way, you can read out that the two proteins did in fact interact. You also have the fragment complementation essay. Here you instead take the two proteins X and Y and you fuse them to the two halves of a split enzyme. If the two proteins interact, they will reconstitute the enzyme activity by bringing together the two halves of the split enzyme. Both of these essays measure direct protein interactions, but they have a higher error rate. Another very popular type of experiment is APMS, affinity purification followed by mass spectrometry. For this, it's important in the affinity step to get a very clean pull-down. And for this reason, people often use the tandem affinity purification approach. Here, you have a construct where the protein X, the bait, is fused to both a calmodulin binding protein, a teft cleavage site and protein A. This allows you to first do a pull-down with an antibody against protein A, then cleave off the construct, do a secondary pull-down using calmodulin, and then another washing step. This gives you very clean pull-downs that allows you to then do mass spectrometry and identify po-complex interactions. The problem is that this essay does not allow you to measure transient interactions since they will long have been lost in the two washing steps. Finally, there's proximity labeling. Most popular is the BioID approach. What you have is a reactive group that you can fuse to other proteins with an enzyme. So you fuse the enzyme to the protein of interest, and any protein that comes nearby will now get the BioID tag attached to them as a form of post-translational modification. This allows you to subsequently measure every protein that has been nearby the protein X. This allows you to identify transient interactions very well. The problem is that you get rather noisy data since sometimes proteins randomly get very close to each other, and they will be tagged too. So how do you go about integrating these experiments, and why would you? First off, you need to be aware that these different essays are not directly comparable. It's much better to think of them as being complementary, since they work well for differentiated proteins as shown in this plot. Here, I'm comparing yeast to a hybrid, fragment complementation assay, and APMS. As you can see, the fragment complementation assay works much better for transmembrane proteins than the two others, whereas the two others work better for nuclear proteins. There's thus no such thing as the perfect assay to get a good overview of all the interactions in a cell you have to combine several. To do that, you generally need to turn everything into binary interaction networks, which is trivial for most assays, like yeast to a hybrid and fragment complementation assays, since they are inherently binary. But it is a bit tricky when you want to do this for APMS data. Imagine that you have two complexes. You have ABC, and you have the alternative complex AB prime C. If you do a pull-down using A as the bait, you can represent it in two ways. The spoke representation, where you show that A pulls down B, B prime, and C, or the matrix representation, in which you link all four proteins to each other. Which one is the better? Ultimately, it's a trade-off between sensitivity and specificity. If you use the spoke representation, you leave out some correct interactions, for example between B and C and B prime and C. But if you use the matrix representation, you infer a wrong interaction between B and B prime. So it's not that one is better than the other, you have to pick what is more important for you. This is all I want to say about interaction experiments. If you want to also learn, or you can predict interactions, have a look at this presentation. Thanks for your attention.