| abstract
| - @jflat06> i really want to get centroid mode in the game... [02:16:59] centroid mode? [02:18:22] <@jflat06> centroid mode makes things a lot faster at the cost of precision [02:18:34] <@jflat06> makes early-game much more managable [02:18:50] sounds nice :) [02:19:06] I might have considered submitting that as a request before... [02:19:07] just what we need [02:19:18] ...but didn't because I thought that this would be too much insider information [02:19:41] <@jflat06> the new alignment panel uses it [02:19:43] <@jflat06> in dev-prev [02:19:56] <@jflat06> the 'score' in the title bar calculates using centroid mode [02:20:22] <@jflat06> which basically means it treats sidechains as a single large 'atom' in a single position [02:20:28] jflat, centroid mode won't have the hiding contribution will it? [02:20:28] <@MikeCassidyToo> new dev to dl?? [02:20:49] <@jflat06> as of yesterday, yeah mike [02:20:49] Brow42: I doubt it [02:20:50] oh, it will? I knows sidechain orienation? [02:21:01] <@MikeCassidyToo> I got that one [02:21:10] <@jflat06> there was another new one today [02:21:14] <@jflat06> that resolves some autosave issues [02:21:20] After all, what centroid mode does is simplifying the side chain to one giant sphere [02:21:27] <@jflat06> yeah [02:21:27] So an entirely different score function has to be used [02:21:28] <@jflat06> basically [02:21:49] well, I just want it to know if I have the oranges and blues on the right side of the sheet [02:22:00] The side chain terms would have to be replaed with a lumped term that captures van der waals interactions, I think [02:22:02] <@jflat06> it approximates each AA sidechain by a single large atom [02:22:05] <@jflat06> so you never have to shake [02:22:14] that explains why it is call centroid mode [02:22:19] <@jflat06> yep [02:22:19] <@jflat06> lol [02:22:29] <@jflat06> but yeah, you have to use a different score function [02:22:31] I thought it replaced SEGMENTS with a large atom [02:22:32] Obviously that's a huge loss of information for long side chains [02:22:34] <@jflat06> or the game will crash... [02:22:35] E.g. Lysine [02:22:47] <@jflat06> not segments [02:22:51] <@jflat06> the backbone atoms are all there [02:22:54] <@jflat06> just the sidechain atoms [02:23:11] cool, this sounds like a good improvement for early game as you said [02:23:28] yeah, or arginine [02:24:11] <@jflat06> generally when someone uses rosetta on an extended chain, they'll use centroid mode, and then refine it later in fullatom mode [02:24:18] well, it will be interesting to see if centroid mode works in practice [02:24:27] <@jflat06> it works for the baker lab [02:24:28] <@jflat06> lol [02:24:57] <@jflat06> and i believe local minimize and the pull tool actually convert portions of the protein to centroid [02:25:03] <@jflat06> err local wiggle [02:25:20] How does that work? [02:26:12] <@jflat06> it just changes the residues in the copy of the protein it's working on temporarily while it's working, then copies the backbone atoms back when its done [02:26:14] <@jflat06> i think [02:26:41] <@jflat06> but yeah, it's pretty cool with the alignment tool [02:26:49] <@jflat06> you can immediately see the feedback of say, flipping a sheet [02:27:06] <@jflat06> if you shift by one, it gets horrible, and shifting another time will bump it right back up [02:32:36] jflat: I always like to try a new feature, so please let us try this new mode out, and we can give feedback on how it works [02:32:56] <@jflat06> if it was that simple, i would have done it long ago [02:33:06] <@jflat06> there's quite a bit of work that needs to be done to get it working in foldit [02:33:32] <@jflat06> the entire game assumes you have a full-atom pose everywhere [02:33:43] <@jflat06> it wouldn't take kindly to plugging a centroid pose in instead [02:33:44] <@jflat06> lol [02:34:35] How does centroid mode handle clashing/steric hindrance? [02:35:31] <@jflat06> clashing is all the same, dont know about the hindrance. [02:36:45] sounds like it's just one big ball [02:36:55] <@jflat06> yeah [02:37:03] <@jflat06> from my understanding [02:37:08] <@jflat06> seems dead useful though [02:37:09] so it won't pack right [02:37:24] *correctly [02:37:25] <@jflat06> well, it will pack 'approximately' [02:37:32] <@jflat06> but running repack on it will do nothing [02:37:33] <@jflat06> lol [02:37:37] <@jflat06> (aka shake) [02:37:49] <@jflat06> which is the reason why shaking does nothing to the score in the alignment panel [02:38:14] so the size of the ball depends on the size of the sidechain? That would sort of explain how it would avoid clashes. [02:38:16] Is the centroid score calculated the same way as well? [02:38:30] Namely, Foldit score = 8000 - (Energy score) * 10? [02:39:01] <@jflat06> @tokens, yeah. but you still get clashes if the atoms are too close [02:39:12] <@jflat06> and yes, it's the same function [02:39:22] <@jflat06> though with different score terms in the underlying energy function
|
![[RDF Data]](/fct/images/sw-rdf-blue.png)