Computing Chemicals and Crystals

In order to design useful new compounds, we need to know exactly what structure any new chemical will take. Computer models, combined with more traditional crystallography, are...
14 August 2011

Interview with 

Dr Graeme Day, Cambridge University


In order to design useful new compounds, we need to know exactly what structure any new chemical will take.  Computer models, combined with more traditional crystallography, are leading the way in predicting how any given molecule will arrange itself...

Helen -  We've just heard how the structure of proteins is vital for their function, but does the same thing apply to other types of molecules we might want to set out designing for ourselves?

Graeme -  Yes.  Well I'm looking particularly at much smaller molecules than proteins, and we're not just interested in the molecular structure itself but in the material structure that we get when we crystallise a molecule.

Helen -  So that's when you put lots of molecules together to make a bigger chunk of that stuff essentially.

Graeme -  To make a solid state structure, and the most stable solid form of molecules, is to make a crystal structure where the structure repeats periodically.

Helen -  So presumably, the structure within the molecules will then tell you about how it's going to form a solid structure when you've put lots of molecules together.

Graeme -  Well, a lot of what we're trying to find out is how the molecular structure relates to the crystalline structure that you end up with when you crystallise a molecule. This will have to do with how the functional groups, sort of the arrangement of atoms on a molecule, line up against each other, and the favourable ways that these functional groups want to line up against each other will dictate the way that the molecules arrange when they meet each other in solution and crystallise.

Helen -  Right.  So how are you actually going about modelling and predicting these structures of molecules, and the crystal structures when they come together?

Graeme -  Okay, so I've been working on this problem of crystal structure prediction for quite a few years now.  We're trying to develop computational models where we can take a molecule, even before it's synthesiseX-Ray diffractiond, so just the picture of the molecule, say, the chemical diagram, and then we use quantum chemical methods to come up with the three-dimensional structure of that molecule itself.  We put that into a computer then to tell us all the possible ways that that molecule can pack into a crystal structure.

Helen -  So essentially, you're drawing your molecule and from that, you can begin to understand how you think it's going to behave.

Graeme -  Yes.  So we want to then predict how the crystal structure is going to form and that's going to then relate to the properties that we get out of that material.

Helen -  Excellent and what sort of factors do you need to take into account?  I mean, it sounds to me in my head, I can see a computer screen with balls and atoms all arranged into a molecule, and then it's just a case of almost fitting them together like a jigsaw puzzle.  I'm sure it's not that straightforward, but what sort of things do you have to take into account?

Graeme -  Well it is a lot like a jigsaw puzzle actually.  Molecules want to fill space as efficiently as they can when they crystallise, and what they're trying to do when they fill space as efficiently as they can is to find the minimum free energy, so the very lowest energy structure that they can form for a crystal.  So what we're trying to do is build up all the possible ways they can pack together, which ones are filling space best, and then we calculate very accurate energies of all of these possible crystal structures to assess which one is the lowest in free energy, and then we assume that's going to be the most likely one you'll get when you go and make this molecule and crystallise it in the lab.

Helen -  So molecules are lazy in a sense and they want to have this low energy. How do we understand that energy use in terms of - you say there's a low energy form?

Graeme -  Okay, so what that's made of and most of the dominant term in the free energy is the potential energy and that's really just an interaction between those molecules.  So you might have strong interactions like hydrogen bonds, and that has a very low energy itself, that's why that hydrogen bond forms, so the molecules are going to arrange themselves to make as many of those favourable interactions as they can, because they have a lot of this negative potential energy which leads to a very low free energy.  You can, under very specific conditions, try to make metastable crystal structures.  They might have different properties and then you're trying to derive it into something which is not the lowest free energy state and then you'll get a metastable structure.

Helen -  Okay and can you also work this backwards?  Can you say, "Okay, this is the structure I want.  I can now go about actually creating that"?

Graeme -  Yes.  Well, we have some sort of empirical rules like these hydrogen bonds.  We know what functional groups want to line up against each other.  We're trying to put this more in a very reliable computational model so that we can try to say, "I want this end structure, this type of arrangement of molecules in my crystals, how can I draw a molecule that's going to lead to that?"  And we can then do experiments in the computer to tell us which are the most likely molecules that are going to give us that structure.

Helen -  And once you can predict a structure, can you take it a step further and predict the properties that are going to come out of that structure?  Is it going to be that easy to go from one to the other?

Graeme -  Well this is the grand aim.  We don't want to just come up with a nice beautiful looking structure of a crystal.  Really, what we want to design its other properties and the structure of the crystal is going to affect things like maybe the colour or the hardness, the solubility, or maybe some interesting properties like charge transport for making semiconductor crystals.  So, some of these properties are easy to predict, some of them are basically just determined by the arrangement of atoms.  Some are quite difficult to predict - things like solubility and there's still a lot of hard work that needs to go into that, but we're plugging away at these different properties so that we can put all these together and try to design properties from scratch in the computer.


Add a comment