 So based on the success of identifying genes and how they are similar to other genes, that means I should just be able to throw some bifuramatics at the problem and then we're done right. Not so fast, because the problem is I don't only want to identify things that are evolutionary related or how they've duplicated genes, I want to be able to go from the sequence to the structure, which is the topic of this class. And this is way harder, at least for me. It's not obvious how to translate these. Here too, it's hard to convey exactly how much this field has developed. So I did the same thing here. I stole an old slide for myself. Some 15 years ago, I had this in a presentation in a bioinformatics class. And at the time, I argued that secondary structure prediction, that was of medium difficulty, but we could probably get some 80% certainty or so. Homology modeling that I'm going to come back to in a second, that was fairly easy, but maybe, maybe not. And Abinitio, that was impossible. There was even a joke at one of the mailing lists of a program called Amber, which is a friendly competitor to our code Gromax. There was a poor student that at some point asked, what does Abinitio stand for? And some joker of the senior developers chimed in and said, it's Latin for doesn't work. Secondary structure prediction, it's not true. That is a medium difficulty anymore. This is trivial. We, there are many methods here that get kind of 90% accuracy here, and that's roughly as good as experiments are because there will always be some minor deviations in loops or so. This is not even a problem anybody's working on anymore. It's literally cracked. And that's amazing. I will have you show you just a slide or two about it for historical accuracy, but it's really not worth spending much time on. Here we just use the server and we're happy the problem is solved. Homology modeling is an interesting problem. So homology modeling means that for these sequences where we have roughly 30% identity, we know that there is another related protein that they do have a common ancestor. And for at least one of those related proteins, there is a structure known. In that case, I can try to build the structure for my new sequence based on the structure of the old sequence. Because again, if the sequences are very similar, it's just a matter of lining them up against each other. This is no longer fairly easy. This is easy and very reliable. We use it all the time. It's used in the pharma industry. A fair number of the proteins in the SARS-CoV-2 genome were modeled instantly with homology modeling and they've been of excellent quality. I'm not saying that you should never determine a structure. There are many good reasons for a determinate structure. But if there is a homologue of known structure, just use homology modeling and you're going to get a fairly good structure, I hope, within a couple of hours. And then we have this interesting final line. Even a few years ago, as recently as three or four years ago, I would say that it's pretty darned hard. There are some programs that have made some progress and I'm going to show you a couple of them. Since two years ago, roughly, I would argue that this problem has been solved. I would never have guessed this but it has been solved largely with deep learning methods and I'm going to show you just a few slides of it because I suspect that this is going to be the future of the field.