Levinthal:
I can understand your frustration here– even as someone who has been playing this game for 1.5 years, I still see many cases where very creative methods are needed to restructure the strands/helices the way I want them to. The main reason the problem you're observing exists is that, by default, the program ensures that the residues stay together (i.e. the end points are fixed unless you're working with the termini). In other words, the reason that you can't bend a structure in a certain direction in some cases is because it's physically impossible.
Now, since we're working with a simulated structure in computational terms, a possible technical workaround is to relax the constraints during the intermediate steps and only require that the final structure fits the constraints. This is actually a feature in Rosetta (the software suite used for protein structure prediction/design that's also adopted in a simplified form in Foldit)– "cut points" may be inserted between which two residues to temporarily allow them to separate (and be processed individually) and then rejoined later. The challenging part, however, is implementing this in Foldit as well.
==> See, the crux of the issue is really that what we want to do with the protein as humans can be difficult to program. After all, you'll have to first find a way to translate what you want to do into operations involving residues and atoms (translate or rotate in a certain direction while fulfilling certain constraints) AND ensure that the code works the way it's intended… which is easier said than done (and I'm saying this from personal experience, having worked with the C++ code of Rosetta in the past). Foldit, after all, is a computer program– the protein puzzles are not produced as a physical toy, a model kit that one can play with bare hands.
.
.
.
.
- That being said, here are a few hints that you can try regarding your specific questions:
- In general, if a structure won't rebuild, consider setting several of its neighboring residues to loop temporarily and including them in the process; adding bands might also help. For your information, what rebuilding does is called "fragment insertion" in technical jargon– basically copying short segments from a database of actual proteins based on the secondary structure and trying to fit them into the space that's available given the end point restrictions. Hence, the most common reasons that rebuilding doesn't achieve what you want even after many tries are (a) the segment you're rebuilding is too long (i.e. too many variations to try from), or (b) the resulting structure is too energetically unfavorable to be acceptable, or (c) no structure from real proteins can reasonably fit into the space given. If it's (a), rebuilding shorter segments one at a time might work, while if it's (b) the solution might be dialing down clashing importance temporarily.
==> But if it's (c), you're in trouble because that means extra pre-processing is required. If you're rebuilding something into a helix but can't, it's probably because there is too much space between the two end points; with beta strands, it's probably the opposite. And if you're trying to untwist a beta strand but can't, your best bet is probably restructuring it the hard way– namely, going into full-atom mode, banding the needed atoms, and then rebuilding 3-4 residues at a time along its length. And if that doesn't work, try freezing everything except for those 3-4 residues, applying bands of strength 10 (first apply regular bands, then create a simple recipe that sets all bands to strength 10 automatically), and then wiggling. The backbone score will most likely drop by the thousands, but at least it will twist in the direction you want it too (remember, you can always fix the backbone later through local wiggling, etc.) For best results, clashing importance could be dropped to 0.10 or even 0.01 as necessary.
.
.
- One common reason that the protein falls apart is that it's too energetically unfavorable. Once again, this is why wiggling at clashing importance 0.01 + using bands to keep certain parts of the structures together is so important– because it allows you to first fix the worst clashes partially. [For your information, wiggling is known as "gradient-based minimization" in the technical jargon. In other words, the program will move pieces around based on the direction(s) that gives it the greatest increase of score relative to the extent of transposition made– which, unfortunately, sometimes means exploding the protein and missing the best structure. Wiggling at a lower clashing importance may solve the problem by first turning a -10000 clash into a -500 clash, which is much less likely to be so overwhelming that the rest of the protein must be sacrificed to fix the most energetically-unfavorable parts.
==> If that still doesn't work, you want want to freeze most of the protein and wiggle them (again, starting at clashing importance 0.01 if necessary) one part at a time. For the record, this issue is actually far more likely to occur when you're playing with guide-matching puzzles or those that involve threading, and the chance approaches one as the number of residues increase. If you're not trying to match a specific guide, extreme clashing is less likely to occur– but when it does, remember that you don't have to follow the default option of doing a global wiggle with clashing importance set to 1.0. The option to decrease clashing importance is there for a reason– in fact, that tool was vital in my attempt to swap two adjacent sheets in Puzzle 165, which is mentioned on Page 17 of the supplementary information section of the recent Nature article ( http://www.nature.com/nature/journal/v466/n7307/extref/nature09304-s1.pdf ).
- Bottom line: when what you're trying to do doesn't work, experiment with creative use of the tools. Unfortunately, this could mean a steep learning curve as such tactics often need to be discovered on your own (because they would be too complex for intro puzzles)…
