1 reply [Last post]
Joined: 04/24/2014
Groups: None

This was an impromptu feedback discussion on the new overhauled Remix tool. It was posted by popular demand here. You can discuss it further right here in this thread though. :) (And we encourage you to.)

Susume's picture
User offline. Last seen 3 days 15 hours ago. Offline
Joined: 10/02/2011
more chat about remix

We had some more chat about remix today - sorry my formatting is not as pretty as jflat's.

(7:17:56 AM) Susume2: if remix only returns a small number of results, it partly defeats the "deep" idea of DRW, which is to compare a large number of results
(12:04:08 PM) jflat06: for those interested, i've posted yesterday's discussion on the remix tool, as people asked
(12:04:11 PM) jflat06: http://fold.it/portal/node/2002353
(12:08:50 PM) MikeLewis: thanks
(12:09:27 PM) Susume2: thanks jflat
(12:10:08 PM) Susume2: does the remix tool pay attention to which aa is in each spot? like will it only pick a good proline position for a proline?
(12:12:58 PM) jflat06: in general it doesn't pay attention to sequence, with a few exceptions
(12:13:07 PM) jflat06: proline is one of them
(12:13:24 PM) jflat06: 'beta branched residues' (I, V, T) are another
(12:13:34 PM) Susume2: looks like maybe glycine is one too?
(12:13:39 PM) jflat06: and it also restricts glycine space for non-glycine residues
(12:13:44 PM) Susume2: ok
(12:13:46 PM) Susume2: thanks
(12:13:53 PM) jflat06: yup
(12:45:53 PM) Susume2: if remixed areas are coming back with lower ideality scores, a shake and wiggle to test each one may not be sufficient to see if they are actually better than what you started with
(12:46:31 PM) jflat06: they might not be, no
(12:47:13 PM) jflat06: i found that the ideality scores were within similar bounds to the unideality produced by rebuild, though
(12:47:21 PM) jflat06: at least in the test cases i was running
(12:47:22 PM) MikeCassidytoo: Idealism is always suspect in proteins and politics
(12:47:41 PM) Susume2: hehe
(12:47:58 PM) MikeCassidytoo: ;-)
(12:48:09 PM) Susume2: I have yet to find a remix that improves the overall score after shake/wiggle, but I've only tried a few areas
(12:48:11 PM) jflat06: if we start seeing bad artifacts, we might have to filter the results
(12:48:51 PM) jflat06: well, if your protein was optimized before, i wouldn't expect it to beat anything with just a shake and wiggle
(12:48:55 PM) MikeCassidytoo: That is what I have found. Random places seems to give pts after shaking and wiggling
(12:48:58 PM) jflat06: does rebuild beat them in that case?
(12:49:28 PM) Susume2: it's still early game, lots of play left in it, and it has some bad rama dots after a long-segment run of DRW that I am trying to fix
(12:50:51 PM) MikeCassidytoo: Are ram dots like voids you can not get ri of voids and therenwill always be some bad rama dots?
(12:51:22 PM) jflat06: i'm pretty sure you should be able to get rid of all of them
(12:51:47 PM) jflat06: though it may not always be advantagous
(12:53:38 PM) jflat06: do remixed sections tend to not have bad rama plots?
(12:55:04 PM) Susume2: most of them have good rama, though not all
(12:55:39 PM) Susume2: again, have only tried a few areas
(12:56:55 PM) Susume2: sometimes remix will put a seg into a rama place that would be good for most aas but is bad for the aa it put there
(12:58:00 PM) jflat06: ahh, yeah, that will happen
(12:58:30 PM) jflat06: my hope is to eventually have remix be able to change the size of the loop in design puzzles, too
(12:58:41 PM) Susume2: and its idea of where glysinces can go is broader than the colored areas we have to glycine
(12:58:46 PM) jflat06: so you'll get a lot more results of various sizes
(12:59:37 PM) Susume2: *for glycine
(1:03:32 PM) jflat06: it will be interesting to see if/how the improved remix affects fragment quality
(1:05:15 PM) Susume2: if I use remix on the exact same pose is it supposed to deliver the results in the same order?
(1:05:25 PM) jflat06: it should, yes
(1:05:32 PM) jflat06: however, repeatedly using it may not
(1:05:42 PM) jflat06: but if you were to undo, it should do the same thing
(1:05:47 PM) Susume2: ok
(1:06:06 PM) jflat06: should help my sanity with testing a bit
(1:06:21 PM) Susume2: trying to come up with a routine that will let me put all the results I like into the undo graph so I cam work back through them testing each one
(1:06:42 PM) jflat06: hmm
(1:06:56 PM) jflat06: perhaps i can add an extra button to "save to undo graph"
(1:07:07 PM) Susume2: that would be excellent
(1:07:20 PM) jflat06: although using the undo graph for that purpose is a little hacky
(1:07:30 PM) jflat06: i had considered being able to have a save quickbar within the tool
(1:07:51 PM) jflat06: but that wouldn't allow you to play around with them
(1:07:53 PM) Susume2: hacky to you, but matches how i use it with rebuild - lay down several results, then work back through them
(1:08:13 PM) jflat06: yeah, it follows the workflow that is established by existing tools
(1:08:29 PM) jflat06: but if we're trying to improve the game, we want to try and think about how to do things better
(1:08:53 PM) jflat06: that said, that might be another battle outside this scope for now
(1:11:00 PM) jflat06: do the quicksave hotkeys work while you're cycling through remix results?
(1:11:09 PM) Susume2: sadly, no
(1:11:59 PM) Susume2: also, that only gives you 3 slots, you have to run a little script if you want to save to or load from slots 4-99
(1:14:10 PM) jflat06: are shift-4-9 used for anything?
(1:14:58 PM) MikeCassidytoo: Is the first seg always on an axis of the rama map?
(1:15:59 PM) Susume2: so far my best workflow is, save beginning pose to slot 1, run remix, stop on pose 2 (putting it in undo graph), restore slot 1, run remix, stop on pose 3, restore slot 1, repeat for remaining remix poses (skipping any i don't like)
(1:16:19 PM) Susume2: shift-9 is used for secondary structs
(1:16:37 PM) Susume2: yes the first seg is on one axis, the last seg is on the other axis
(1:18:08 PM) jflat06: hmm maybe secondary structure save should be on 0
(1:18:15 PM) jflat06: and 1-9 should be all quicksave slots
(1:18:38 PM) jflat06: as far as i can tell, 0 isn't used for anything else
(1:23:53 PM) Susume2: 1-9 as quicksaves sounds great to me; of course some people will not like moving the secondary structs
(1:46:15 PM) alcor29: If you do 1-8 you wont have to shift the 9 used currently by SSs.
(1:52:51 PM) jflat06: you could, but 0 is a much more natural place to put it. it isn't in the sequence of quicksave numbers 1,2,3,4... and so there's less of an expectation that it will behave like a normal quicksave slot
(1:53:10 PM) jflat06: that, and you're limiting yourself to one less quicksave slot by trying to preserve behavior
(2:03:09 PM) alcor29: ok
(2:05:25 PM) jflat06: i'll try to figure out if there's a way to enable quicksaving while remix is running
(2:05:52 PM) jflat06: it looks difficult with our current code though


Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, Boehringer Ingelheim, RosettaCommons