Which one is better?
Do you see the difference between them?
My impression is that rebuild could be faster. Not sure about it.
Rebuild is disabled on most design puzzles. For those, it's really fast, but doesn't actually do anything.
Aside from that, remix is newer, and the interface is better. Remix saves its results in quicksave slots, making it easy for recipes to retrieve them. For manual use, Remix also has a good user interface, letting you look at each result, and save the ones you interested in quicksave slots 1-8. (I'd recommend using the selection interface, which lets you select exactly the segment you want to remix.)
Rebuild is not so user friendly manually, it basically keeps trying different shapes, and you have to stop it if you happen to see something you like. In a recipe, if you want to look at 10 rebuilds, the technique is to rebuild with a cycle count of 1, then check the result. Next, rebuild with a cycle count of 2, and check the result, then 3 and check, and so on. If I recall correctly, this is how Tvdl enhanced DRW 3.1.1 works. That doesn't seem very efficient, but most likely rebuild will never get a makeover to work like remix.
One area where remix is weak is the segments near the ends of the protein. It usually says "no replacement found" for anything near the end. That's something to do with the end segments not having a peptide bond on both ends, making it impossible to calculate one of the two backbone angles (phi and psi). It would be nice if remix could just ignore the missing angle, but that's probably not in the cards.
See this intro to remix from when it was new. It explains the difference between remix and rebuild in detail.
I found to get extra 200 points with 'Tvdl enhanced DRW' with the segments 5-3 when the script 'Tvdl DRemixW' seem to be stuck with the same parameters on a small puzzle.
So rebuild could be more efficient (by points), but that is the matter of a deeper research.
I often use the recipe Loop rebuild 9.0b
which can be run using remix or rebuild.
In this recipe, running with rebuild
seems to gain points faster, so if I have
the choice, I usually run it with rebuild
and set its rebuild length to a multiple
of 3 (12 first for coarse runs, 9 after
that, 6 after that, and 3 last for fine
runs). When I use remix, I set the rebuild
length to 6 at first and to 4 later on.
Back in Puzzle 1475, for an earlier version
of this recipe, I tried remix with many
different rebuild lengths and compiled
the chart below:
rlen skips ways files:
3 35 83 i3,i7,j7
4 7 82 i4
5 5 81 i5,i6
6 4 80 j5,j6
7 3 79 i8
8 4 78 i9
9 3 77 i10
10 76 76 i11
11 75 75 i12
12 74 74 j1
As you can see, with the rebuild length
(rlen) set to 3 or 10-12, there were many
skips (remix attempts that the recipe
skipped for some reason). The ways column
is how many different lists of segments it
tried to remix. My protein had 85 segments
for these runs, so rlen 3 did 83 groups like
1-3 2-4 3-5 to 83-85 while rlen 12 did 74
groups like 1-12 2-13 3-14 to 74-85.
If you have a machine that can run multiple Foldit
clients at the same time (I have several Windows
laptops that can do this), I encourage you to
try some experiments like races between clients.
You can start several clients at the same time
from the same structure with the same score, but
on different clients use different recipes or
the same recipe with different input variables.
After a couple hours, usually one client is
clearly ahead of the others. At this point, copy
the best solution from its client to the other
clients and then try the experiment again. This
is a good way to learn which recipes work best
in certain situations and can let multiple slow
clients compete better with one fast client.
Remix was introduced to help us design proteins
more likely to fold in the lab. I think its
library of protein fragments is less extensive
than rebuild's library to help this happen.