The joy of seminars

Like most researchers, I go to a couple of seminars a week (on average). I take copious notes and never look at them again. So, I thought what I would do is flick through those notes and pick out a few choice morsels that caught my attention for some reason or other:

1. Bridget Trogden (Mercer University, GA, USA)

I'm only briefly going to summarise the work, because that wasn't what caught my eye about this seminar. Bridget works on ligand binding to proteins related to pregnancy related malaria, going through steps of molecular modeling, synthesis of ligand mimics and binding assays.

What was really interesting is that the university is predominantly teaching-based and almost all of the work had been done by undergraduates and masters students. This meant the project had been chopped up into bite-size chunks that were manageable over whatever time period each student was assigned in the lab. In addition, the work was carefully chosen to be in an uncompetitive, but significant area, as was carried out at a somewhat slower pace than at a more research-oriented university.

I hadn't come across anything like this before and it was a nice reminder that the way I experience research isn't the only way to do it.

2. Scott Woodley (UCL, London)

This time it was the science, or rather the maths, that prompted me to remember the seminar. It was about global optimization of the structure of zinc oxide, particularly the location of oxygen vacancies.

Let me give you some background. Global optimization is the problem of finding the absolute lowest energy value in a system. You might like to think of the analogy of finding the lowest point on an 18 hole golf course. There's lots of bunkers which, although low, aren't the single lowest point. When you get to hundreds or thousands of dimensions, this becomes a very tricky problem. There's essentially three approaches: gradient, Monte Carlo and genetic. I'm  not going to explain them, although if you want to know, put a comment and I'll write a follow up post.

The crux really is that proponents of each method sings its praises. In this talk, Scott was proudly explaining that you can't beat a good genetic algorithm. For my PhD, my statistics supervisor assured me there was no better way to characterise your solution than with a Bayesian Markov Chain Monte Carlo (MCMC) algorithm, but in my viva a mathematician assured me that this was way too over-complicated and a simple gradient method was more than sufficient for my (admittedly fairly simple) model.

This all leaves me confused - which to choose? I think it's a question of context. For a simple, quick estimate, start with a gradient method. For a medium-sized number of variables, try MCMC and if your model is really complicated then Scott's probably right. I'm still trying to work this one out, though...

3.  Brian Austen (St George's medical school, London)

Brian (or should I be more deferential and say Professor Austen) presented research into Alzheimers, in particular the detection of beta amyloid oligomers using biotinylated antibodies (for the non-specialist, read this as 'attached a big sign to a small molecules that says "Look at me, look at me!" so you can detect it using well established methods).

Interesting work, and drew my attention for a couple of reasons. Firstly, this happens to be my brother's turf and I was hoping to impress him with something new I'd learned from someone eminent in the field. Secondly, there was a healthy dose of what I call realism (others call scepticism) in the approach to the science. The amyloid hypothesis, that build up of beta amyloid causes Alzheimers, has long been accepted, but recent research is starting to question it. Prof Austen accepted this, without particularly expressing an assertion either way. Also, he accepted and clearly stated that the next major hurdle in his work is delivery across the blood-brain barrier.

This seemed a really balanced approach to me. Clearly identify the problems, then be willing to work your ass off to solve them. Also, when data contradict an established result, don't hold too tightly to your conclusions, but be ready to challenge them and drop them if need be.

 4. Stefan Auer (University of Leeds)
This was a talk with great science. The motivation was the phase diagram of a protein: what temperature and concentrations does a protein form alpha helix, beta sheet and/or fibril structures at?

Computer based prediction of protein folding is a very tricky problem, and an extremely important one. Protein fold structure determines function to a very large extent. Stefan has built on work by Nguyen and Hall, which is a model without most of the properties of the protein, just keeping the ones considered relevant to folding. It turns out that a finite thickness backbone can be good enough to get an accurate fold.

Tweaking the protein interaction parameters gives some really nice videos mimicking protein aggregation. The conclusion was that phase diagrams are affected by the ratio of hydrophobic interactions to inter-chain hydrogen bonds. All in all, work I was told to be nearly impossible a few years ago.

Experiences of a doctoral training centre

The stereotype of the PhD model is that some aging professor gets grant money out of thin air, offers some magic beans to a wide-eyed recent graduate and three years later the jaded, bitter PhD candidate emerges, moaning they haven't seen their supervisor for two and a half years, thesis in hand to take their viva. This system seems to have sufficed for many a year, but now the EPSRC and others are starting to offer something different.

This is where my  PhD, finished about a year ago, comes in. I became a part of MOAC, one of the EPSRCs first doctoral training centres (DTCs). The idea is that you take in a cohort of students, teach them together for a year, ending with an MSc, then send them off to various parts of the university to do a PhD for three (ish) years.

Does it work? In my opinion it has a lot of strengths and is a good complement to the traditional system. Here's my advice on how to run a good DTC:

1.  Get the right people through the door. Able, motivated, independent students. They will not belong inside a cosy, traditional department and are likely to be presented with unfamiliar problems, academic and otherwise - it's not all about raw intelligence, but also about being adventurous.
2. Good administration. The difference between DTCs that I've seen succeed and fail is the administrator. Bad administration means the students feel permanently lost and like they don't belong anywhere.
3. Leadership. Alison Rodger at Warwick drove through the creation of MOAC and got the science departments and university management behind the idea. It simply wouldn't have happened without her. A balance of shrewd politics and pastoral attention seems to be the best mix.

What's different about a DTC?  Well, one thing to consider from the MOAC model is that when choosing a PhD subject the students come with money and can approach an advisor with a project idea. This puts the student in quite a powerful position. Turning down a project doesn't have anything like the consequences of the traditional system where you're wondering 'where will the next offer come from'. Good or bad? I'm not sure, but it does not seem to lead to silly projects. At the same time there must have been a few good projects that failed to get a student.

There's just a few points to contemplate. For some more info on MOAC, see http://www2.warwick.ac.uk/fac/sci/moac/ and for a list of all the EPSRCs DTCs: http://www.epsrc.ac.uk/funding/students/centres/Pages/byuni.aspx