I hope this make sense.

Because we are over searching some areas, is it possible to statistically
work out the spread to the random numbers assigned per heuristic?

If so you could just give out statistically superior random numbers (for
that heuristic), and not have to go to properly spaced ab initio for a
better search of the space (which I believe would be better but not ideal).

Any news on a global optimisiation? This would entail that R@H trickle back
and take into account the fact that some people are not always joined up to
the net, and of course the capped Internet traffic of some ISP's.

It's maybe too big of a jump to have CPU cores talking to each other, LAN's
talking to each other etc...

A question: do AA change shape, I mean as a discreet entity, within a
protein, or just the torsion on the C(a) atom?

Hugo, perhaps you could change your subject line to something more about the random number seeds and search space?

While we're here, I'd appreciate understanding exactly what the random starting point for each model means and how they are used. I mean I know they are highly unique across users and models, and I know if two users hit upon the same random starting point, then they should also crunch through to he same result. But I don't understand what it really MEANS. Does it identify a sequence of torsion angles for the various AAs in the backbone? Or just point to a specific AA to start the searching from?

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

Feet1st, I must be missing something, as I don't seem to be able to edit the subject line, any ideas :?

Ah got it, just need to post then edit that post :)

I hope this make sense.

Because we are over searching some areas, is it possible to statistically
work out the spread to the random numbers assigned per heuristic?

If so you could just give out statistically superior random numbers (for
that heuristic), and not have to go to properly spaced ab initio for a
better search of the space (which I believe would be better but not ideal).

Any news on a global optimisiation? This would entail that R@H trickle back
and take into account the fact that some people are not always joined up to
the net, and of course the capped Internet traffic of some ISP's.

It's maybe too big of a jump to have CPU cores talking to each other, LAN's
talking to each other etc...

A question: do AA change shape, I mean as a discreet entity, within a
protein, or just the torsion on the C(a) atom?

we are testing several things along these lines. we identify torsion angles that are enriched in very low energy structures, and in further rounds biasing sampling towards these angles rather than sampling the very wide initial distribution. we also identify features that occur overly frequently, such as pairings between nearby beta strands, which are formed all too easily during folding, and penalize these features in a subset of the next round of work units.

the starting random number seed determines the initial set of torsion angles each amino acid in the protein gets. it is extremely unlikely that any two users will get the same seed, as the number of distinct numbers produced by our random number generator is many orders of magnitude greater than the number of work units being sent out.

---

the starting random number seed determines the initial set of torsion angles each amino acid in the protein gets. it is extremely unlikely that any two users will get the same seed, as the number of distinct numbers produced by our random number generator is many orders of magnitude greater than the number of work units being sent out.

Ahh! I finally get it, thanks Dr D.B.
I was having a very hard time trying to work out just what you were up to with the random numbers.

we are testing several things along these lines. we identify torsion angles that are enriched in very low energy structures, and in further rounds biasing sampling towards these angles rather than sampling the very wide initial distribution. we also identify features that occur overly frequently, such as pairings between nearby beta strands, which are formed all too easily during folding, and penalize these features in a subset of the next round of work units.

That sounds like it will be very promising indeed, much better than trying to work out patterns in random number sequances :)
edit