This page is just some notes on guide-matching and alignment, with the focus on manual methods and intuition. Perturbation methods for score have been well-covered elsewhere in the wiki.
- The basic character of the protein is determined by the alignment and identity of the sheets and helices.
- Rebuilds of the backbone early on are crucial in determining the peak score of the fold.
- The protein becomes comfortable with early rebuild choices and unhappy if you try to make big rebuild changes late on, punishing you with a reduced peak score.
- The protein remembers local wiggles which greatly increase mojo, this memory can only be erased by rebuilding the segments which were local wiggled.
As a result of these assumptions, I don't apply any local wiggles at all until the endgame, even though they could give me easy middlegame points, because I want to keep my options open and avoid local wiggle high mojo making local energy minima of the protein tough to bounce out of. Also, I restart and match a variant solution as described further on, to bed in major shape changes as early as possible. These assumptions and choices may seem drastic to you and it is fine for them to be questioned. Other people's intuition may take them in different directions. The more different ideas and approaches the better.
Matching native and evolution guidesEdit
Start by aligning the guide on the most central sheet. Band sidechains at the ends of this sheet to the guide. I use the first bend in the sidechain as the band origin. This simply encourages the backbone to twist so the sidechains are in the right places. You can band to the segment by eye if there is a glycine segment with nothing sticking out. I don't usually apply bands to both segments and sidechains anywhere or put bands at both ends of a long sidechain because I want to have the least number of bands possible, but sometimes those extra bands could encourage the protein to relax closer to the guide so keep them in reserve. These initial bands should not be too long as you aligned on the sheet. Put a band on a sidechain in the centre where it is closest to the guide. Now (global) wiggle to shrink the bands to points. Start banding the areas on either side of your initial sheet. At some point you may get a problem with some strand of backbone trying to clash with the strand you are banding. Just use a long band to pull it right out of the way. You can delete this band when the clashing area is safely banded. Now work your way out from the centre, banding a sidechain every so often. For a small protein you might want to band every sidechain straight away to get the alignment right. For large proteins I suggest using as few bands as you can get away with to get the basic shape so the main sheets and helices are matched to the guide, even if the loops are not at all well-matched. The more bands that are not points, the slower it will take to stabilise with wiggle. In the early stages I also use the Blue Fuse recipe to stabilise quickly at a higher score. As you get more bands shrinking to points, save and try deleting some bands to give more freedom to work on sections that are not well-matched. If this makes things worse after wiggle, restore the save. At some point you probably will get a section where the bands don't shrink to points with wiggle. If this is just a loop and doesn't interfere with banding the overall shape, you can ignore it for now until the overall shape is good and most bands are points. With luck this will subtly improve the awkward section by getting the ends of it short-banded. But if it includes an important sheet or helix, you have to deal with it straight away. The first thing to try is to save, then try pulling to get a sweet spot where the long bands have been as much shortened as you can manage. Don't worry about your score taking a nosedive at this point, with atom clashes appearing. Just wiggle, wait for the protein to bounce around violently and stabilise, and see if the band lengths have been reduced. Don't worry if you lost score, if you reduced the band lengths that is progress. If not you can restore your save. The next thing to try is to rebuild for shape the section with long bands, to reduce the band length. I use the selection interface to select the section to rebuild and open the undo graph and use the rebuild tool, just waiting until the rebuild pose has many of the bands shortened. I use a graph length of 60, which is two pages long. It may be difficult to spot the best pose with the selection on whilst the rebuild tool is working. So after it has created just under two pages of rebuild poses, cancel it and cancel the selection. Then go back to the beginning of the rebuilding on the undo graph. Now you can work at your own pace, rotating the view of the protein to see how long the bands are. Just find one with as many short bands as you can. It works best when your section is about three to five segments and the segments at both ends have fairly short bands. Then you can just fix the problem with one rebuild. But for longer sections you may have to break up the section into five segment pieces and do several rebuilds, and have the added complication of the ends of your piece not both having short bands. If you can get the overall shape and alignment of the most important, large sheets and helices right, that is more of a priority than trying to fix a long loop. This is because the loop sections may drift from the guide anyway even in the highest scoring and most compact and pleasing folds.
Drifting from the GuideEdit
When you have got the main sheets and helices close to the guide, you can save, delete the bands, stabilise with wiggle and Blue Fuse, and just run a rebuild script. I use Walkin Rebuild Prot, which avoids any local wiggling. This lets you easily see how much drifting from the guide is happening and where it is worst. But the result is only useful if it preserved your main structures and didn't mess up your helices or sheets. If it kept your main structures in place and in shape, you can now see the drifting sections, where you can be creative with rebuilding for shape to generate some variant folds. But the protein will change a lot as you move the score up, even if the changes are subtle and hard to see. My assumption about early rebuilds determining the score peak is then an issue. The obvious problem is that you don't know at the start which rebuilds for shape might work out, but you can easily try them at the end. But the protein then resists these late major changes and limits its peak score. A solution is to save the late rebuild variations, with their lower peak scores, restart and load the variant as a guide. You can match it just as you did the native, but now you have a fixed choice for the drifting sections and you shape rebuild them to closely match as early as you can. The protein becomes comfortable with the new shape early on and may reward you with a higher peak score. But the highest peak is going to depend on how good your choices of reshaping the drifting sections were. That's where you can be creative and use your folding experience and intuition.
The position of one or both tails of the protein can be an important early choice, if the tail drifts and there are several choices of where to place it, perhaps held in place by strong hydrogen bonds, perhaps floating free. Non-local hydrogen bonds, if strong enough to persist and affect the overall shape of the protein, are also important early choices. But many bonds will break and reform differently and so are not so important at the start.
The rebuild tool is undocumented, so some info I have picked up may be useful. But take it with a grain of salt as it is just my guesswork.
Sometimes the rebuild tool gets "stuck", failing to offer you even one new pose after a minute or two of running (so on the undo graph nothing moved from your position before the rebuild was run). I would advise giving it 30 seconds to a minute, as sometimes it is apparently stuck but then gives you new poses which may be very good. Getting stuck may be because you have structure (helix or sheet) inside your selection and the tool is trying to restrict to rebuilds that are in keeping with your structure, so having a more difficult task than with just loops. In this case, you can save your structure with CTRL SHIFT 9, then convert your selection for the rebuild to loops, then run rebuild again, and this should get it working again - though not necessarily with useful poses! You can then use CTRL 9 to restore your structure.
The rebuild tool (in the selection interface, at least) appears to work with bands. If you have a selection with sidechains (say) banded to the guide, but the bands won't shrink to points with wiggle, or the sweet spot pull, this may indicate a need to rebuild. In the selection interface with your section to rebuild selected, run rebuild with your bands still active. If you are lucky, the rebuild tool will grind away, appearing to be stuck, but then give you poses which match your bands more closely! Once again, the tool may simply get stuck and offer you no new pose. This may be because your bands are already quite short, or the end segments of your selection have long bands, making it difficult to get a matching rebuild that connects with these segments, or you just gave the tool too much for it to cope with. I usually only give it selections of three to seven segments and try to get the end segments fairly short banded when using bands. Again, keeping in structure may make the rebuild tool get stuck, so converting to loops may help it match your bands.
Obviously, each rebuild initially drops your score and needs to be stabilised with some clashing/shake/wiggle combination, and perhaps a blue fuse. I consider getting the shape right to be well worth risking a drop in score that I may soon recoup. Particularly well chosen manual rebuilds can add a hundred points to your score after stabilising, and any successful one can add five to twenty points, assuming a middle game score. But at first your rebuild efforts may fail to recoup lost score - it takes time to develop the shape intuition that allows you to be successful so I advise persistence as manual shape rebuilding is your chance to be creative and do something differently from script rebuilding.