I've been watching the graphics for rosetta, and
comparing the "accepted" "low energy" and "native"
foldings. The native are abstractly pretty - they
have a recognizable quality to their compactness.
Almost like Cristmastree ornaments. The analogy is
good - they fit neatly inside a minimal radius sphere.
"Low energy" and "accepted" forms don't fit inside so
small a sphere. They often have appendages, or tails
sticking out. They obviously aren't what we're
looking for. A considerable amount of compute power
is devoted to obtaining the energy for these.
Finding the energy is expensive. Determining whether
the form fits inside a certain radius sphere should
be cheap.

If the sphere is small enough, only good candidates for further wiggling will fit inside. Forms that
can't be folded to fit in the sphere can be disposed of without expensive computation.

I conjecture that the "lowest energy" folding is
nearly synonymous with the smallest radius sphere
the thing can be folded to fit into. Even if it's
not (maybe multiple shapes all have the same radius,
but obviously different energies), it's a cheaper
paradigm to search with. The search is therefore a
set of screens - each screen a progressively smaller
sphere. Only when the sphere is "small enough" would
the candidates surviving need a comprehensive search.

Am I making any sense at all? We need to speed this
thing up. As it stands, an exponentially increasing
amount of compute power yields only incrementally
better results. Only an algorithmic change can
reduce that explosion. The algorithm needs to do
more like a human would - I can look at a folding
and sort easily those that might be minimal from
those that clearly are not. I would only play with
the ones that might be.

Stuart

---

An interesting idea, but it will only work for globular proteins. Here is an extract from the introduction of Harold P. Erickson's "Size and Shape of Protein Molecules at the Nanometer Level Determined by Sedimentation, Gel Filtration, and Electron Microscopy":

"Most proteins fold into globular domains. Protein folding is driven largely by the hydrophobic effect, which seeks to minimize contact of the polypeptide with solvent. Most proteins fold into globular domains, which have a minimal surface area... However, some proteins are highly elongated, either as a string of small globular domains or stabilized by specialized structures such as coiled coils or the collagen triple helix."

One of the key goals of Rosetta is to predict the shape of unidentified proteins. If the project team tell the system to ignore results that fall outside of the predicted globular radius then they could indeed speed up finding the globular ones but come up with a completely wrong answer for any elongated or helix-based structures.

---

The main issue I can see with this sort of an approach is that proteins exist in solution and interactions with water play a large role in their final shape, and they do not have any strictly defined overall shape in general. Fibrous proteins, by definition, are not spherical (and are generally insoluble as well) and even for globular proteins (which Rosetta seems to focus on) no particular shape is favored, some have distinctly non-spherical shapes (such as chaperonin), and there's no reason to assume that an arbitrary protein wouldn't have a hydrophillic protrusion sticking out of it somewhere.

I'm not an expert by any means though, only in my first semester of biochem at the moment. Maybe someone more qualified will weigh in?