 So before turning to computers, we can try to do the same thing manually. Remember when we spoke about the ligand-gated ion channels and how anesthetics were influencing them and how the anesthetics were usually targeting the membrane, that they were very hydrophobic. And I showed this slide of modern inhaled anesthetics. It turns out that there is a virtually perfect correlation between how hydrophobic these small molecules are and how efficient they are as an anesthetic. Again, this was originally a phenomenological observation by Meyer and Overton, not really based on a deep insight about the process. But given that, if I now show you one more molecule and it's very, very hydrophobic, and a second new molecule that's very hydrophilic, which one of those two do you think is going to work best as an anesthetic? If you don't know anything else right, you would say the hydrophobic one, and that's of course right. So if there is some sort of, in this case, it's just a simple first-order regression, a fit. But based on that regression, I should be able to find more molecules like the ones I've already seen. Doing this manually this way is completely naive, because every single hydrophobic molecule is not going to be an anesthetic. But with a computer, we can do more advanced things. So assuming that we have one molecule bound, might not be very efficient, but something is bound to a related receptor, even my receptor, then I can start to open a computer and look at this molecule, or maybe it's not even a molecule, maybe it's just a surrounding. So I might be able to say that, this molecule that my colleague already found, it appears to have a hydrogen bond up here, and then another hydrogen bond, let's say it's a donor down there, and then there should be an aromatic ring, and then there should be a plus charge. That's a pattern, right? Maybe he even found two molecules that looked like that way. And then I can ask my colleague, Dr. Johnson, that Dr. Johnson, can you try to find more molecules that have a hydrogen bond donor and a hydrogen bond acceptor, now I forget the other part, a plus charge and an aromatic ring, right? Find molecules that fulfill that pattern. And if you find more of them, let's go into the lab and test them. Today we do this with computers. This is called QSAR, Q-S-A-R. This is a fancy name for something that isn't really so fancy. We're just doing very first simple order correlation. Quantitative, that just means that we're correlating things. Structure activity relationship. So quantitative structure activity relationship is that I try to correlate the structure here with how active it is, and if all the active things have a hydrogen bond, that means I should look for more hydrogen bonds. And then you just put this in a database, and then we iterate, we pick out, say, 10 hits from my database. I take those to the lab, I go in and check them. Out of those 10 hits, I might have, huh, three molecules here appear to have some effect. Let's focus even more narrowly on those three. Maybe those had a very weak hydrogen bond donor up here. So let's, in that case, let's adjust my QSAR rules a bit and look specifically for weak hydrogen bonds. Or maybe some of them had an extra methyl group. So this is just really a way of iterating things to fit things that I've already seen. And today, as I mentioned in bioinformatics, anything that has to do with fitting or so historically, we would do with deep learning today. Today, deep learning is gradually replacing QSAR, but it's very much the same approach. Look at things we've already found that appears to bind to this molecule or other similar molecules, and then use computers to identify more molecules like it. The difference here is that these molecules don't have to exist in databases already. Chemical space is very large. The total size of chemistry space is something 10 to the power of 60. There's probably not a single compound data bank, actual compounds available that's larger than 10 to the power of 6. But if based on my rules, I know it would be really, really, really good to have a weak hydrogen bond donor up here. Well, we can ask our organic chemistry division in the company to go and synthesize that is build such a molecule. That might cost $10,000 or $50,000, but if that is really the only thing that could create a drug for me, it's likely worth it. Maybe I can afford two or three molecules like that, but then they'd better show some good results in this early test, or I'm going to cry.