Thirty years of biomolecular simulation

Image via Wikipedia

Adam Kraut blogs about Wilfred Van Gunsteren’s talk on 30 years of (bio)molecular simulation. This is a topic very dear to my heart since that’s kinda what I am supposed to be good at (or at least was) and it’s also one where I feel that we haven’t been as innovative as we could have.

Physical representations of biological systems are, at least in this writers book, the ultimate form of biological knowledge. Actually figuring out the physics and chemistry of what makes proteins behave the way they do, how they interact is something that is both a very hard problem and one that we would love to solve as a community. According to Wilfred there are four main problems limiting the application of biomolecular simulation today

1. The force field problem
2. The search problem
3. The ensemble problem
4. The experimental problem

You can read Adam’s post for all the details. Personally I think we’ve done a great job with the search and ensemble problems. As our computers have become faster and more powerful, there has been a lot of good work in new search strategies, and we’ve made some progress on the ensemble side as well. The experimental problem is real and a little out of our hands, but there are two areas where I don’t think sufficient advances have been made; the force field problem and the software engineering problem. Now, neither is a trivial problem, but I think we can do better. I like the library-based approach that the SimTK folks are taking, but that’s just a start. I believe those are the areas we’ll make significant advances over the next decade, taking advantage of GPUs, scale out architectures and maybe come up with more relevant force fields as well. Our existing force fields still only describe model compounds, and extrapolate to macroscopic structures. They don’t do a good job of representing macromolecular complexes and while we have come up with some good soft body potentials and reduced representations that help address that problem, but we are still talking about a massive computational problem.

Having said all that, perhaps the most important point in the talk was this

So what if you could simulate every atom in your body for 1 second

The answer: There’s much better things simulation can answer; ask better questions.

That’s great advice, and why we don’t need to solve the protein folding problem in the near term. What we need are better, more usable, representations of biomolecular structure and dynamics, their interactions with other molecules and various other properties that help us answer or propose questions we couldn’t otherwise. Data-driven techniques and mathematical models can address many questions wel, so we don’t have to try and solve every problem using simulation. The simulation community is probably best served by thinking of good ways to represent scales (e.g. Greg Voth’s work), extending upon the kind of work that the SimTK folks are doing, thinking about scalable MD engines, and continuing to think about new wars to sample and explore energy landscapes while extending practical techniques like MM-PBSA that help address real problems.

Related articles by Zemanta

• Paying attention to simulation (mndoci.com)
• Folding Time-Scales w/wo GroEL (proteome-zone.blogspot.com)