Alphatoxin backbone ideality

Started by Skippysk8s

Skippysk8s Lv 1

1440 and the two following alphatoxin puzzles loaded onto my machine (windows 7, main client) with a "frozen" backbone that was not mutatable, but non-ideal. There were approximately 40 red balloons indicating backbone problems
I popped all the balloons I could, then ran both cut and wig length 3 and microidealize length 3. That fixed the back bone, but took about 3 hours since the protein is so big
I will note that every time I tried to add peptide bonds my score dropped
I have heard others did not have this issue. I shared to scientists, and BoKoep and I did discuss on chat. However, seems my machine gets a broken puzzle. BoKoep thinks it is a bug. Many of my team mates had the same problem, as did Retired Michael to my knowledge from vet chat during 1440. Team can share my "solutions" to anyone they want. Not sure why this is occurring.
Puzzle size is so large I am not opening any more alphatoxin at this time. It requires all my CPU to run, so I can't do anything else
Please post what machine and operating system you have if you also have this "feature" to help BoKoep and the team figure out what is going on. Thanks

Hanto Lv 1

more likely we need to know what machine you have and what softwares are on it?. I heard a lot of complaints about mutation regarding 1450 for example, yet none of my machines can concur with negative comments regarding 1450. I have to believe something else is going on with the machines incapable of running 1450. And just what is a peptide bond, don't see it in tools so I may have to consider it to be outside my ability to reproduce.

LociOiling Lv 1

I created the recipe Microidealize 4.0 test ("MI4T") to look at this issue.

On puzzle 1440, MI4T found ideality changes on locked segments, even with its "do locked segments option only" option checked. The gains resulted from wiggling adjacent unlocked segments. The changes were mostly gains, and happened only on the locked segments immediately before or after one of the permanent cutpoints.

It's not clear whether these results are exactly what Skippy is describing, so here's what I did.

In 1440, "show backbone issues" revealed a large number of issues with the starting pose. It appeared that all the issues were in the unlocked sections of the protein, however. Using "ligand specific" coloring, the locked sections appear in gray, and the unlocked sections appear in shades of yellow and red, similar to "score" coloring.

I ran MI4T on the starting pose of 1440 several times with all its new options checked. One nice thing is that the results were repeatable each time, producing exactly the same results. I left "worst first" checked each time, and didn't try to manually correct any backbone issues for these tests.

Using the "do locked segments only" option, along with "include locked in wiggle", and "include unlocked in wiggle", you get gains. With just "do locked segments only" and "include locked in wiggle" checked, there's no gain.

Locked segments can't be selected, either manually (in the selection interface) or in a recipe. So there's no way to idealize or wiggle only the locked segments. MI4T runs very quickly if you leave out the unlocked segments.

As Susume has mentioned, changes to the unlocked part of the protein can affect the subscores of segments in the locked part. The changes to the ideality subscore revealed by MI4T could be just another example. For example, you may see changes to clashing, packing, or hiding in the locked part when you move nearby unlocked segments.

With ideality, it's not clear what's actually changing. The sidechains on the locked section are all locked too. (This is not necessarily true on all puzzles with locked segments.) So it's not the phi and psi angles (our new friends from the Rama map) becoming more ideal. There's also the omega angle, the angle of the peptide bonds between segments. Maybe the omega angle can become better despite the presence of the permanent cutpoints.

Susume Lv 1

A too-long cutpoint reduces the ideality score of the segments on either side of it. A bad-angle cutpoint also reduces the ideality score of the segments on either side of it. The permanent (gray) cutpoints in aflatoxin can have both their length and their angle changed by wiggling just the unlocked segs adjacent to them. This will change the ideality score of both the unlocked and the locked segs adjacent to the cutpoint.

In stick view, a foldit cutpoint does not look like it replaces the peptide bond; it looks like it connects the alpha carbons of the two amino acids - but that may just be how it is drawn. I would hazard a guess that the ideality score is based on the length of the peptide bond (the distance between the final backbone carbon of one AA and the nitrogen of the next) and the omega angle (the dihedral angle about the peptide bond), neither of which corresponds exactly to the cutpoint we see, but both of which get worse when you stretch or bend the cutpoint.

Skippysk8s Lv 1

The view options menu (you may have to turn on full gui) lets you check a box to show peptide bonds. It is on the upper left, most of the way down that list after things like show backbone problems, show sheet bonds, show sidechain bonds, etc.
I have updated my Foldit with the recent releases. Tonight 1440 loaded only with problems on the mutatble ones, whereas my original one was the reverse. I wonder if something changed when jflat made the last couple updates. I can't back it up – I didn't save the prior Foldit

LociOiling Lv 1

Sounds like a good explanation of the "local" changes to ideality of locked segments.

But I've also confirmed another one of Susume's observations – running Microidealize 3.0 on the start pose of 1440, I see ideality changes in two locked segments following an unlocked section. In the first run, this happened at 45-46, 66-67, 87-88, 105-106, 131-132, 166-167, 188-189, 239-240, 259-260, 278-279, and 297-298.

It's not immediately clear why there's a difference. Microidealize 3.0 seems to use random order, where my tests were in "worst first" order. And Microidealize 3.0 does all segments, where my tests only "idealized" the locked segments.