Running Rowan's Conformer Search Workflow

Transcript

Hi, I'm Corin, CEO and co-founder of Rowan. In this video, we're going to look at conformer searches.

First off, what is a conformer search? Well, while some molecules are very rigid and exist only in one shape, most molecules have one or more rotatable bonds and thus can exist in a variety of shapes. The point of a conformer search is to quickly figure out which of these are relevant, are going to be the ones that are low enough in energy to actually happen in the real world—either for binding to an active site, packing in a crystal lattice, or reacting in a transition state—so that we don't have to use our chemical intuition to guess. We can actually use computation to narrow this down intelligently.

It's easiest to show, not just to explain, so let's just try running a conformer search on a simple molecule and talk about what happens. We'll start by clicking the “New conformer search” button from the main page of Rowan and we'll go ahead and draw a 3D structure so we have a molecule to demo. We'll draw n-pentane which is a simple five-carbon alkane and also a laboratory solvent. This is a nice molecule because it's small, so it's relatively tractable (there aren't that many conformers), but there's also a lot of rotatable bonds. Every carbon–carbon bond here can potentially be rotated around, so there are actually a lot of possibilities of different shapes that this molecule can take on.

Now when we're running a conformer search we're faced with several questions: we can choose what solvent we want to use, what method we want to use, and what mode we want to use. So, the solvent actually dictates if we're going to try to model the effect of solvent or if we're content modeling this in the gas phase. For this example, we'll just leave it in the gas phase, although of course it might be more relevant (depending on the application) to model the effect of a solvent. The method asks us which method we want to use to score the output conformers. In this case, we'll stick with the default AIMNet2 neural-network potential, which is fast and quite reasonable for things like this.

And then the mode is the most complex, and that actually asks how we are going to generate and winnow down these conformers. We have some more details if you hover over this little info icon, and what you can see is that there's sort of a lot of things: how many conformers we're going to look for, what program we use to generate them, what level of theory, what energy cutoff. We here at Rowan have put a good amount of time into trying to choose really really good defaults for these so that the users don't have to. Basically what it boils down to is that, depending on your use case, usually rapid is fine. If you want to be very very careful that you've gotten all the conformers careful or meticulous could be better—and if you're trying to quickly search through an entire library where runtime is really, really important, then reckless is probably best. But for this use case, as with most, we'll stick with rapid (which is also the default), and we'll say “submit conformers.”

So what this is doing behind the scenes while it runs is using one of the programs I mentioned earlier. This one will use RDKit to generate lots and lots of potential conformers. and then is going to go through a multi-step process of iteratively optimizing and filtering these so that we remove duplicate conformers and filter out any conformers that are too high in energy to be reasonable—and now it's finished.

What does this output mean? So, we have a list of conformers here in a little table. We have a ∆ energy, a relative energy in kcal/mol, and then we have a weight. So this essentially corresponds to what the energy is relative to the lowest energy conformer. So we can see here that conformer 1 is lowest in energy, it's this sort of slightly bent conformer. Conformer 2 is a fully linear conformer, and that's predicted to be 0.18 kcal/mol higher in energy. So that's almost the same energy, but not quite. Conformer 3 now is quite a bit higher in energy, so that's sort of doubly bent. And we can see that that's 1.2 kcal per mole, higher in energy. and then conformer 4 is this sort of syn-pentane conformer here. So there's this sort of clash between the two ethyl groups around the central carbon atom. And so that makes us, you know, 2.3 kcal per mole higher in energy.

And for those of us who struggle to do logarithms in our head, we can look over here at the weights and see that what's predicted to happen is that conformers 1 and 2 will account for the bulk of the population at room temperature. So conformer 1 will be about 53%, conformer 2 will be about 39%, conformer 3 will have maybe 7%, and the very unhappy conformer 4 will only be 1% of all molecules in solution. So that's what a conformer search looks like.

We can also do some other things. So we can overlay all of these conformers to sort of get them plotted in the same area. And while this is pretty cool—it gives you a sense of sort of the overall flexibility of the ensemble—we can also do some even more interesting things by selecting a couple of the atoms and aligning on that. So let's just see, given that these three atoms we try to align as closely as possible, like what will the overall rest of the structure look like? And we can see that, you know, some of them sort of zigzag off to the side, and others are totally straight. And the opacity here indicates sort of the energy of the conformer, so the darker conformers are lower in energy, whereas the lighter ones are slightly higher in energy.