Best RMS not getting any better?

[KWSN_Millennium2001Guy]

I thought that since we have produced nearly 2 billion extra structures on this protein that there would be some movement in the top RMS structures. There hasn't been a single change in the top 10 structures. Is there something on the server that stopped recording the best structures after the goal was met, or has Murphy's law just prevented us from producing better structures?

Ni! :)

[Paratima]

Dyyryath's current top spot was claimed "Mon Sep 9 13:50:18 2002", long before the 10-billion goal was reached.

5 postings since. I suspect the law of averages, not Murphy, not Howard's devious server.

(Unless it's really devious. ;) )

[KWSN_Millennium2001Guy]

Actually Dyyryath obtained that particular protein sometime after October 2nd. His first top-ten structure was back in September. The ENTIRE top-ten list has not changed since we topped 10 billion structures. I would have expected the tenth-place structure to have been bumped off of the list at least. We have produced 20 percent more structures and some of them should have been as good or better than the 10th placed unit.

[Paratima]

Hmmm. Yes, I see that you sent congrats to Dyy on the 2nd....

Hmmm, again. Two since then, both on Oct. 5th.

Started on Sep 3rd. Hit 10-bill on Oct 9th. It's now the 14th. That's 5 days.

Graph it:
S.....X....XX....XX..........X..X........T

S=Start X=Hits T=Today .=nuttin'

Well within the realm of possibiliy. I still think it's a coincidence.

[FoBoT]

isn't there something like the "law of diminishing returns" ? or something kicking in?

i mean, would it take exponentially more samples to get a lower rms , or is it linear?

seems not to be linear

but i don't know much about all that, i am just a computer geek/moderate stats whore :o

[Paratima]

Moderate? :confused: :rotfl:

[MAD-ness]

The way I understand it, the reason we are doing runs of 10 billion is to find answers to the questions y'all are asking.

However, the impression that I got was that they expected to seeing diminishing returns and 10 billion, let alone 12 billion, is pretty far past the drop-off point, apparently.

That said, an extra 2 billion structures ought to have popped out SOMETHING that was better than one of the top 10 structures. The structures are random, so there is no reason the last 2 billion are any better/worse than any other 2 billion. All it means is that to get a better result than what we have now we would have to greatly increase the number of samples/structures we create.

I think. :(

[Brian the Fist]

What FoBoT said basically. To answer without hurting anyone's brain (I hope), best RMSD vs. # sampled is logarithmic in nature. Thus to improve upon where we were at 1 billion, we needed to sample 10 times as much - 10 billion, to get a relatively noticeably improvement (of about 1A). So similarly now, we wouldn't expect to get below 7A until we reach about 100 billion. Thus it is not surprising to see no improvement in the latest 2 billion.