I have already estabished that at school, out in the the courtyside of Herefordshire, I was quite an enthusiast for science, technology and mathematics. That is not to say I did not enjoy art, graphical design and history, but I excelled at the sciences. For A-Levels (for those in the US that is the equivalent of the last 2 years of high school, but over here we specialize in a few subjects, and for my time that was just 3) I took Physics, Chemistry and Mathematics. My Mum to a degree forced me into doing maths, on the grounds that no matter what I did at university it would come in handy. She was not wrong.
My love of Physics really comes from my childhood obsession with space. I loved how the solar system was, how planets were so different and similar, and at how man had left the confines of this world to explore others. A part of me as a child wanted to be an astronaut, or some form of astronomer. But as high school went on I could see possible choices like engineer or theoretical physicist. However there was chemistry.
Chemistry is a weird science if ever there was one. It sits are the threshold of all the others. Not all scientists wear white lab coats, and not all chemists are the same. Not all work in labs slaving over making new colourful liquids or bubbling, steaming pots of solutions. No there are a lot of boring steps to be taken in the creation of new chemicals. However, there are many forms of chemistry. There are surface chemists, biochemists, bioinorganic chemists, nano scientists, this list being very long.
What captured my imagination in Chemistry, was the links between it and Physics and in turn Mathematics. Quantum Mechanics. This strange area of science, ruled by particle wave waves, and strange physics, is the very science that puts electrons in their place around atoms, and in turn determines how chemistry happens. I was just struck by the beauty of the equations that determined the motion of these particle/waves. And so it was this that made me do Chemistry as a degree. But of course a particular type of chemistry.
Now I had applied to the Univeristy of Manchester Institute for Science and Technology (it was a one of a few such institutes) for a Masters in Chemistry. This was originally a B B C entry requirement, that based on the interview for the course, was reduced to B C C, with a B in Chemistry. That of course was achieved (I got a A in Maths, and two Bs).
But the course I was doing was not normal chemistry. It was Chemistry with Chemical Physics. Chemical Physics I had learnt during my hunt for university courses, was an area of chemistry where computers were used to model and analyse chemicals. It would mean I would learn programming and deal with Quantum Mechanics.
Of course in an ideal world you get to study exactly the way you want to. However, being such a nerd I was one of a few who were doing that exact course. Meaning that in the first few years of the degree I got to study specialized modules in Quantum Mechanics etc.However, Chemistry has a high level of attrition amongst the students. By the later years I was really the only person doing that course. This was an issue as the specialist course were not always an option for me to take due to not enough interest in them. This meant often I was doing other optional courses that were more synthetic in focus. This was and issue as it caused a drop in my overall grade averages. One thing I did learn through team projects is that I disliked doing synthetic chemistry. It would either yeild very small amounts of the desired product, or turn to brown goo. This was why I prefered physical chemistry and theoretical chemistry. it was all the formulas that described the chemical bonds and how molecules move about.
So for my final thesis for my Masters, I did a project on the design of a new, multipolar electrostatic, water model. How can I explain all this concisely? Water is a the most fundamental of all molecules. It is the medium for life, and is essential for the understanding for many other important chemical systems and physical properties (like how ice freezes). So what exactly was I doing?
Above is the way water molecules organise in the liquid. This structure is constantly shifting in the liquid, but becomes rigid in ice.
Water has been modelled since the start of Computational Chemistry, back in the 70s. Water models assumed a number of things. Water molecules are rigid (molecules are anything but rigid). Water molecules don't break bonds (this is a massive simplification - water molecules are constantly exchanging hydrogen atoms and making and breaking hydrogen bonds - these being weak interactions between the water oxygen atoms and the hydrogen atoms on another water molecule. Even if we models did do this, they assumed that hydrogen atoms move like normal atoms, but in fact hydrogen atoms are so small and light they move in non-Newtonian ways i.e. Quantum Mechanically).
Above is a diagram to show how electrons fill atomic shells. The number of electrons in a outer shell determines the chemistry. Atoms react and bond so that they complete a shell either by losing electrons or gaining them. For example, Carbon, has 4 out electrons. It reacts to form 4 bonds. In each bond it shares one electron from itself and another from the bonding partner atom. Thus in total Carbon has 8 electrons in total. A complete shell.
Water molecules can be described using points charges placed on the atoms. Oxygen atoms carry a partial negative charge, while the hydrogen atoms carry partial positive charges (this explains the above mentioned hydrogen bonds). These partial charges simplify the true distribution of electrons about the water molecules. The old models assume these charges never. However, these charge distributions do change, in response to bonds being made and broken, and in fact changes to the local environment of the molecules. This is called polarization (something I will get back to later).
Above a typical water model. It has the bond lengths and geometry. Note that water has a triangular shape. The toal charge of the molecule is 0. But the oxygen atom has a partial negative charge, and the hydrogens have partial positive charges.
So what was my model. My model used a more realistic representation of the electrons and where they are located, something called a electron density. These are 3D representations of the charge density and you can imagine the analogue with respect to say pressure of temperature.Above is the molecule, imidazole, and the gradient vector field of its electron density. Note the field lines are the lines tha end at the dots (atoms). The isobars represent lines containing equal electron density. The thick curved lines are interatomic boundaries. Note how they curve. This means atoms are not round things when in molecules. They deform each other. The image is the same for computational determination as it is if you measured the same thing by x-ray diffraction. In fact the computer calculated version is more accurate. The above image is similar to the previous. This time for two water molecules. One water molecule, on the right, lies in the place of the 2D plot. The othe is at right angles to it, with the hydrogen atoms sitcking out of the image. Note how the left water molecule oxygen atom deforms the hydrogen atom of the right hand water molecule.
A 3D representation of the atomis of three water molecules within a cluster of 21 water molecules. Red volumes/atoms are oxygen atoms, white are hydrogen atoms. The solid atoms belong to the central water molecule of the cluster, while the two neighbouring molecules have transparent wire-framed atoms.
This project not only saw me learn more about quantum mechanics and use such programs to generate data using the equations of quantum mechanics, but I also learnt about programming, Fortran, in order program the models and modify them so that using Newtonian equations of motion I could test if the water models recovered the expect structures of water that have been previously been measured using X-ray diffraction.
The work for this revealed some interesting results which I would then make use of in my PhD with the same group. The PhD was offered to me so long as I got a 2.1. Thank fuck I did.
Getting a PhD was a life changing experience. First of all having funding and money is good. Especially when you go from three grand a year to twelve. My PhD involved learning more programming and the fundamentals of AI, in particular neural networks. The new project was 'The design of a novel polarizable water model trained on ab initio electron densities'.
What hell does that all mean?
Let's go back to the old work. Remember I said the model assumed that the charge densities don't change, and that was a simplification? Well this new model of mine woud address that. The neural networks are computer algorithms that can learn things from the data presented to them. So what data is that?
Ab initio is the latin for 'first principles' i.e. quantum mechanics. I generated thousands (and that takes some time) of quantum data for various water clusters i.e. 2-6 water molecules in different arrangements where one molecule is surrounded by the rest. The data for these clusters shows that the electron density is distorted due to the placement of the water molecules. Why? Remember I said that the atoms have partial charges? Well that means that water molecules interact in such a way that the partial charges either push (negatively charge atoms do this) or push neighbouring electron density in other molecules. This distortion of electron density is known as polarization (I hope you note that a lot of these terms can be looked up on wikipedia).
So this means that each water molecule, and its electron density, are unique to the environment and organisation of that environment i.e. what stuff is about it and how they are pointing at each other. A neural network can be trained to related the positions and relative orientations to the electron densities found for these examples. In effect the neural network can predict during the simulation of water the electron densities, and in effect allow for the water molecules to be polarized.
Above is a basic neural network. They are an analogue to how brain neurons work. The circles, nodes, pass numbers along (left to right). These numbers are multiplied by things called weights i.e. how important a the number is, and used to calculate an output. A neural network 'learns' by modifying these weights so that once it has been trained to predict the output for some test examples, it can then be used to predict the output for other sets of inputs representing the other variations you wish to use.
Are we still following? Well this work is now being applied to a model for peptides (short chains of amino acids that if you make big enough can curl up and become proteins) and for water with ions (salt water is a good start).
So then I finished at the University of Manchester (a merger of UMIST and the Victoria University of Manchester) and have almost finished a postdoc at Warwick University. Here I have been developing models for spin crossover compounds.
OK. So there are these types of atoms in the periodic table called transtition metals. These metals are called so because they can easily under go a transition from one oxidation state to another. That means they can lose a variable number of electrons when forming different complexes. For example iron can happily form compounds where in some it has lost 2 electrons, and in some 3, and in others even more. This means that in the two states it prefers different compound geometries i.e. what shapes it forms when binding to other atoms, it also has different colours in the two states, which in turn are further modified by the atoms it is bound to. For example rust is red because it is iron in the 3+ (i.e. lost three electrons) state. This is why our blood is also red when oxygenated. Really, go look up transition metals and see why they do so much stuff and are so important to life and science.
Above shows the outer shell electron structure of iron. Electrons are arrows. The lines are orbitals. Transition Metals break all the rules. Sothey have a 4s shell that holds 2 electrons and a 3d shell that can hold 10 electrons. Both shells are similar in energy. When iron is oxidised, it looses the the 4s electrons first (becoming a 2+ state). It will then loose one of the pair electrons in the 3d shell to form the 3+ state.
Now the other thing that transition metals can do is occupy different spin states. This means that while the oxidation state is the same, the electrons in the outer shell of the atom (the outer shell of the atom determines the chemistry of an atom) can be forced to change their arrangement. In turn this means that they can favour different geometries with the atoms to which they are bound. It can also mean they can be trapped in either spin state. (Spin is a property that electrons have. It is either up or down. Electrons can only be in the same orbital if they are of opposite spin. Now per orbital there are two electrons of each spin. This is stable. But pairing electrons decreases stability because electrons are negatively charged. It's like putting two north poles against each other. So then it is also favoured to have electrons spread out, one per orbital if possible. So there can be a number of ways to spread the electrons.
One is where as many electrons are paired up - low spin, and one is where as many are not paired up - high spin).
Above is the 3d shell for the iron in the 2+ state. What you need to know is that when iron binds with atoms you will find some of the orbitals (those lines the electrons are on) are higher in energy than others. Now here is the trick. You gain stability with the electrons in the lower energy orbitals. But you loose energy pairing electrons. So the you can spread them out (LS being low spin as the spins are all cancelled out, HS being high spin where there are more up spin than down spin). But that means putting electrons in less stable, higher energy orbitals. So there is a clever balance here that deteremines if the high spin or low spin state is preffered. It depends on how unstable i.e. how much higher in energy the upper orbitals are. If the pay off is not enough then LS state is preffered. Of course this energy difference, and thus preference can be modified by changing what iron is bound to.
What does this change of spin state allow for? Well spin states can be switched between if the material adsorbs a gas, or is heated, or is compressed, or is hit by a laser light. What can we do with this? The spin state can be used as a form of switch, like in memory in hard drives. Or perhaps as sensors for gases. They can even be used for optics.
So what am I doing for this. Well many of these models need settings to be determined for the functions that model these systems. Now this not trivial when there are 30 or more that need to be found so that the parameters can be used to model the compounds in both low and high spin states. Now, to find these, since there are many combinations, I have been using genetic algorithms (a way of varying bit string representations of the parameters) to search the parameter space to fit the models.
What makes it harder is that the fitting of the parameters must achieve two goals. The first being good energy predicitions for the test compounds, and the other being good recovery of the compound geometries. These two goals are in competition i.e. you can fit the models to get one really good while getting shit results in the other. This issue is know as multi-objective fitting. This is now going to be applied to a number of problems, and will in future be used for some other things.
My future work in Bochum, Germany, returns to my PhD work or neural networks, and I will be using it to simulate transition metal catalytic surfaces. This means I am drawing upon my old skills and pushing the work further forward since my old work and this new work are comparable and can be combined.
But why do we do this?
In 2003, the cost of developing a new drug was estimated at $800 million, with a predicted 7.4% increase in costs per year, the development of a new drug will now require around around $1 billion.
Typically, it takes over a decade for a drug to be brought to the market because only a couple of potential drugs out of 10,000 make it to the market. Moreover, it can be difficult recoup the money put into the research and the drug may be recalled when it makes it to the patient population. Subsequently, drug development and production needs to become more efficient. This can be achieved through the use of computational chemistry Computers have become ever cheaper and faster. It is, therefore, now feasible to run moderate sized simulations on a commercially available desktop computer. By using the computational tools available, and developing new computational approaches, drug design can be made more efficient and successful.
So I guess that means what I do should hopefully help save lives, or save the world. No really.
So there we go. My Speak Out With My Geek Out about Chemistry