Eric Drexler is one of the smarter people I know. We don’t always agree. In fact, we disagree a lot, but he is always an interesting read. In a recent post on the Macromolecular Modeling Blog (cross-posted to his blog) Eric writes about macromolecular modeling in the context of one of his favorite subjects, molecular systems engineering. It’s an interesting post and worth a read.
I won’t go into the molecular nanosystems part, cause that’s the bit that I don’t always agree with. What I would like to talk about are some of the more general paradigms and needs Eric puts forth. He talks about what is needed to explore design, widening the design scope, and increasing the design scale. Things that jumped out at me include the funnel from high throughput to high quality, an open extensible design exploration framework, extensible force fields, physics-based energy functions, coarse-grained models, and integration with finer grained models. These are all problems relevant to modeling macromolecular systems, both from the design perspective as well as simulating their behavior and predicting interactions with other macromolecules and drugs.
Multi-scale modeling is the holy grail of molecular modeling and simulation in my mind. Being able to represent dynamic systems (or not so dynamic ones), their properties and interactions at a variety of scales is required to represent the physical behavior of biological systems. I always imagine this framework with an orchestration engine manipulating models at different levels of detail and maintaining information transfer between them, optimizing force fields and properties on the fly. It’s a tough problem and no one has solved it yet, although there are some attempts to do so in the nanotechnology/materials science world. Some day, hopefully, we will get there.
Eric also postulates the following
The fundamental difference between scientific modeling and engineering design is that in a design process, the physical system isn’t a given entity, but is instead found through exploration guided by a functional objective. This has far-reaching consequences.
Here is where I have some concerns, mostly around the fact that while our functional objective might be defined, the methods to get there right now are terribly inaccurate, especially across scales. So do we really know what we are getting, and to make these systems biologically useful is going to require more detail than we have right now. Having said that, we can take the rules we have learned from structure prediction and inverse folding and apply them to the design problem, much like David Baker’s group has done and over time we’ll get there.
The other interesting question he asks is about when a model is good enough, which also speaks to the issue above. I agree that design doesn’t need to solve scientific problems, but a design needs to be scientifically accurate. I am not so sure if a model fails for science it works for design, unless the design only serves as an initial seed to experimental design.
I will add my obligatory warning now. I am much more vested in understanding biological processes and the molecular recognition problem. I am also fascinated with designing biological systems, but over the years have come to the conclusion that biology is much too complex for realistic design of systems given current knowledge. We need to solve the scientific problems first, then we can really focus on the design problems.
Thanks for your commentary on my post and perspective on molecular and multiscale modeling. About the question of whether we're ready to undertake design problems, this of course depends on the problem.
One benchmark is that structural DNA nanotechnology, at least in its DNA origami form, has typically been done without using any computational modeling at all, so sometimes the bar for the adequacy of modeling is set very low. Another benchmark for modeling requirements is in protein engineering, where reliable methodologies for designing stable protein folds have preceded those for predicting the folds of stable proteins.
Growing research in molecular engineering will boost support for science-centered macromolecular modeling, and we already know enough to build better design tools. Unknowns will always be prominent in science, since the unknown is what science is about. Design work reverses the figure and ground, focusing on what we do know and trying to leverage it in creative ways. The remaining unknowns and modeling defects are why cutting edge work always involves a large measure of trial and error experimentation, and why it helps to focus and drive scientific exploration.
A related question is what physics can tell us about objectives that out of reach because of the limitations of current fabrication capabilities. The answer is somewhat paradoxical, and the specifics of what can be understood are both very limited and very substantial. This sort of study involves design methodologies again, but from a different angle, which I discuss here: http://metamodern.com/2009/02/25/making-vs-mode...
Modeling molecular systems
Eric Drexler is one of the smarter people I know. We don’t always agree. In fact, we disagree a lot, but he is always an interesting read. In a recent post on the Macromolecular Modeling Blog (cross-posted to his blog) Eric writes about macromolecular modeling in the context of one of his favorite subjects, molecular systems engineering. It’s an interesting post and worth a read.
I won’t go into the molecular nanosystems part, cause that’s the bit that I don’t always agree with. What I would like to talk about are some of the more general paradigms and needs Eric puts forth. He talks about what is needed to explore design, widening the design scope, and increasing the design scale. Things that jumped out at me include the funnel from high throughput to high quality, an open extensible design exploration framework, extensible force fields, physics-based energy functions, coarse-grained models, and integration with finer grained models. These are all problems relevant to modeling macromolecular systems, both from the design perspective as well as simulating their behavior and predicting interactions with other macromolecules and drugs.
Multi-scale modeling is the holy grail of molecular modeling and simulation in my mind. Being able to represent dynamic systems (or not so dynamic ones), their properties and interactions at a variety of scales is required to represent the physical behavior of biological systems. I always imagine this framework with an orchestration engine manipulating models at different levels of detail and maintaining information transfer between them, optimizing force fields and properties on the fly. It’s a tough problem and no one has solved it yet, although there are some attempts to do so in the nanotechnology/materials science world. Some day, hopefully, we will get there.
Eric also postulates the following
Here is where I have some concerns, mostly around the fact that while our functional objective might be defined, the methods to get there right now are terribly inaccurate, especially across scales. So do we really know what we are getting, and to make these systems biologically useful is going to require more detail than we have right now. Having said that, we can take the rules we have learned from structure prediction and inverse folding and apply them to the design problem, much like David Baker’s group has done and over time we’ll get there.
The other interesting question he asks is about when a model is good enough, which also speaks to the issue above. I agree that design doesn’t need to solve scientific problems, but a design needs to be scientifically accurate. I am not so sure if a model fails for science it works for design, unless the design only serves as an initial seed to experimental design.
I will add my obligatory warning now. I am much more vested in understanding biological processes and the molecular recognition problem. I am also fascinated with designing biological systems, but over the years have come to the conclusion that biology is much too complex for realistic design of systems given current knowledge. We need to solve the scientific problems first, then we can really focus on the design problems.