23 replies [Last post]
Joined: 12/07/2007
Groups: Street Smarts

I thought it would be good to start a thread with some basic tips for people to use in playing the game. It is certainly much more enjoyable when you have a feel for what you are doing. I don't really scientifically know what I'm doing. But I've been studying Rosetta for a long time. And spiralling those native sturctures around trying to see patterns from one protein to the next. And I seem to be scoring well, so I thought I would share some about things that seem to be working for me. I hope others will do the same and post their ideas here. Some combination of all of them will ultimately be a better approach.


 

In fact, let's make it a standing rule that if you take the top score when a puzzle is closed, you have to post a tip to this thread.

Joined: 12/07/2007
Groups: Street Smarts
Line spacing is broken again

I'll add some tips once the paragraph spacing problems on these msg boards are fixed.

feet1st's picture
User offline. Last seen 7 years 12 weeks ago. Offline
Joined: 01/25/2008
Groups: None
Align the hydrophobic sidechains

Firstly the sidechains are always colored and the backbone that they  are attached to is gray. There are two views of the sidechains. The view is controlled from the File pop-up, then select "view options". There  is a score view, where coloration depicts how "happy" a sidechain is  with his position in the protein. And there is the hydrophobic view.


I always start a puzzle basically by pulling it apart, and then
displaying it in the hydrophobic view. This depicts sidechains with two
different preferences. The hydrophobic sidechains are colored green. These  prefer (i.e. score better) when pointing inward towards the greenish  voids. The blue ones are hydrophilic. These prefer to point towards the  outside, which is presumed to be basically surrounded by water (hence the  blue). The suffix -phobic indicates an aversion to something, in this case water, and -philic indicates an attraction. So, the more blue you get on the outside, and the more green you get on  the inside, the better your score will grow. And you will also notice  rotational forces as you very gradually drag the backbone, or even just  click down on it and do not move the mouse. If you carefully note the  surrounding sidechains, you will see the rotation that is occuring is generally drawing more blue sidechains outward, and more green inward.  Since both are generally on the backbone near each other, it is  sometimes difficult to see the results of these forces, because they may be  cancelling each-other out, or may be surpassed by other forces like a collision, or near collision. You will also want to consider hydrophobia carefully when selecting which portion of the backbone to drag. If you intend to pull it inward, then best to grab a green segment. If you intend to pull it outward, then best to grab a blue. This is because the point you grab basically is an anchor point where the rest of the backbone is flexing to accomodate. So as you grab a blue segment and pull outward, you tend to slightly rotate the neighboring segments in that direction as well (pulling a green around to point more directly inward at the same time hopefully).

feet1st's picture
User offline. Last seen 7 years 12 weeks ago. Offline
Joined: 01/25/2008
Groups: None
VIDEO on hydrophobic sidechains

Hugothehermit and I are working together to bring you videos to illustrait some of the tips of the day. Sometimes a picture is worth a thousand words. Through the videos, you can also get a feel for how others use the tools in the game to achieve the scores they get.

This first video illustraits the importance of understanding the hydrophobic and hydrophilic sidechains. In the version of the game captured in this video, the hydrophboic sidechains are shown as green. These want to be away from all of the water molecules that surround the protein, and so you will want to arrange them to point inwards into the center of the protein. The blue are hydrophilic. They like the water, and will tend to be pulled around to point outward in the form of the protein found in nature.

In the current version of the game, you click the checkbox to glow hydrophobic sidechains, and they are presented as glowing yellow for the hydrophobic, and glowing blue for hydrophilic rather then the green and blue coloring shown in the video. Either way, it's easy to remember, blue LIKES water and you will score better if you find ways to get them to point outward.

Hugo created the video, and I'm hosting it on my website here:

http://www.violetoaks.com/hugovideo/start.swf.htm

It is 22MB. Just let the whole thing download, then click the play button.

In just 131 seconds Hugo shows you how to get a score that would put you at 7th place on top scorers list!

Hugo shows you how to get a great score! ...but forgot to turn on his microphone. So, what you hear in this video is actually his brain as it solves the problem.

Note how the protein folds fairly easily in the directions he's pulling it. This is due to the properties of the hydrophobic sidechains. If you are finding it difficult to pull, it is likely because you are trying to pull it in a direction that is not natural for this series of amino acids.

Also note how he rotates the protein around and pulls from a different direction to get the action he's looking for.

Hugo used a tool called CamStudio 
to capture his screen and produce a swf file that many people can view in their browser without any additional plug-ins.

If you would like me to host a video for you, you can send me a PM here on the game website, or on Rosetta. On the game website, you identify my user ID (feet1st), on Rosetta, they use the user ID number (44890). I will send you back instructions on how to upload videos up to 150MB.

Joined: 12/07/2007
Groups: Street Smarts
This was a test, I can

This was a test, I can downscale and add audio to it, though I'm not sure that a puzzle 17 is really useful?

RosettaMod also needs a mention as he tested my first video capture, and we talked about it, though he never said if my Australian accent was understandable :)

Adrien also told me that the video capture was (paraphrasing)  recorded in a higher resolution the down-sized to 640*480 (I'm relying on my bad memory here take with a grain of salt)

RosettaMod also metioned a different video capture programme (without going through my e-mails I can't tell you it's name) that may be better. Also that the flash file could be played in quicktime.

I havn't played Foldit for a while as RL is getting in the way, and for the forseeable future this will continue.

All files, video, audio etc...  I release as  Copyleft, GPL, or GFDL, what ever you prefer.

feet1st's picture
User offline. Last seen 7 years 12 weeks ago. Offline
Joined: 01/25/2008
Groups: None
Radical movements you did not intend

Have you ever been starting a drag or doing a WiggleBB and suddenly had the whole protein flip flop around like a fish on land? For brief moments there is more red on the screen then gray? This is not a bug! Think of all these residues and sidechains as tiny magnets. And what happens when you get a bunch of magnets together? They suddenly snap into position. Often in a way you did not intend.


This tends to occur when you attempt to drag at or near what I call an anchor point. This is part of why I pull apart the puzzle when I start. Pull it out in to a more-or-less straight horizontal line. Now grab the backbone about a third of the way in from the left, and lift it upward. Do you see the anchor? The point after which your drag had no effect? Go ahead, really pull it hard! Now do the same on the right. Actually, it has only twisting effects, but you can't lift that specific residue (segment in the backbone).

This anchor point gives you something to push against when tweeking your model. It can help you make coils flex out or compress more tightly together. But if you happen to click within about 2 residues on either side of the anchor you have a lot of leverage against the rest of the backbone. So, depending on how things are folded up, sometimes you push segments of the backbone right through each other.


If you click and hold on the residues at and near the anchor, you can see some pretty interesting effects when the protein has been straightened out. It will start to coil in a mannar that it finds natural. Try to follow and foster this natural coil and your model will tend to be more stable and score better.


Sometimes the radical movement is simply due to stresses in your structure that you cannot see. If you hold a drag click on other residues in the protein (those further from the anchor), it tends to relieve stresses in the structure. Even though you don't move the mouse to do a drag, the structure (often) moves. If you have made a major swing of an arm of the protein, you will want to do this click and hold at several points around the pivot. This will help balance out the stresses along that portion of the backbone, and thus avoid the sudden movements during the wiggle backbone.


Doing a Wiggle-all before a WiggleBB can also help relieve stresses.


Also, if you have collisions (indicated by red) in your model, these are high stress points. I find it is generally best to relieve those stresses yourself with the hold of the drag click rather then to let the wBB straighten them out.

Joined: 11/12/2007
amazing tip

Feet1st,

This is an amazing tip! I'm learning a lot. I can't wait to put your technique in action. Btw, have you played with the pivot tool? It allows you to change the pivot point for a drag. (We're thinkig of folding user-defined pivot points into the general drag tool... what do you think?)

Adrien

feet1st's picture
User offline. Last seen 7 years 12 weeks ago. Offline
Joined: 01/25/2008
Groups: None
Now sure what you mean

Are you saying allowing several hinges to be created?

I'm not sure why... but the neat thing about the existing pivot test tool is that the sidechains seem to flex, and you can get the sidechains to align nicely. Just like the anchor point I referred to.

For example, open puzzle 18 to it's initial state (if you have saved a puzzle, you can open puzzle 18, save a copy of what you've got, then tell the game to switch to puzzle 18, and it will restart you at the initial state.

Go to the hydrophobic view. Then go to the end with the ring that's curled around. Select the test Pivot tool. Now inward from the ring there are 4 blues in a row. Right click on the first green after that (it's residue 69 if you click it in the drag view). Now pull on that ring and uncurl that end. Did you see all the sidechains in the whole puzzle move? I don't know what that's about, but other then the anchor and the pivot tool, I don't know if any other way to do that. They seem to be in a permanent wiggle state or something. They will actually bend out of your way if you drag something up near a sidechain.

...now, click on the residue we placed the pivot point in and hold it down (we're in the pivot tool now). Hold it for 5 or 10 seconds, don't move the mouse at all. Did you see that whole end of the backbone flex around?? Now... ready for something cool?? Click on the next segment inward, away from the ring end, and hold THAT down for 20 or 30 seconds! Once it stops moving much, move to the next segment inward and hold a click on that.

What seems to happen is that the backbone balances out all the stresses and pulls all the hydrophobic sidechains to one side and the hydrophilic to the other... just as you would want. And from there you can go about doing your folding.

I don't know of any other way to get what is often such an exact alignment on all the sidechains. And I suspect that is a good measure of why many folks are having trouble getting higher scores.

feet1st's picture
User offline. Last seen 7 years 12 weeks ago. Offline
Joined: 01/25/2008
Groups: None
Unanchored??

Oh my God! P18 just came unhinged! The double ring is no longer an anchor. I've not seen that happen before. It seems the pin tool, even is nothing is pinned, releases the normal anchor. When I went back to drag, now it is anchored again.

Joined: 12/07/2007
Groups: Street Smarts
Tweeking

Once you are satisfied with your model, and feel you have aligned the hydrophobic and hydrophilic sidechains as best as you can, you can pull more points out of your model (and more voids out of it) if you do a lot of tweeking.

Watch your score closely and if it drops, then use the "undo" function to get back to the point where you started. Now drag, very slightly, a portion of the protein, try to pick a portion near the exterior. Push or pull it. Now do a wiggle back bone action. If you gain more then one
 point, try the same thing on the same portion of the backbone again. Do this around many points of your structure. Do this as you click and hold to relieve tensions. Sometimes you have to click several portions of the protein before each wiggle to make any progress on the points.

You will see that generally the wiggle backbone puts things back where they were... so why does it score better? I believe it is because the scoring system is so accurate, that it detects the very minute differences in how the resulting structure is formed. Your drag and the wiggle caused a slight difference in rotation, and this brought a greater attraction between your hydrophobic sidechains, and so it scores better.

Joined: 12/07/2007
Groups: Street Smarts
Much ado about undo/redo

As you are tweeking, it is often a real timesaver if you can see where things are heading and move them further in that direction. After you've done several tweeking moves, and completed a wiggle, bring up the "undo" menu and click on the prior peak in your score. As you do so, notice which direction things are moving. Now you can move to the right and click to return to where you were. Again notice the areas of the structure that are moving. Perhaps these areas will bring you more points quickly with further tweeking.

If you can't clearly see the changes that the wiggles achieved, you might want to rotate the protein around to another viewing angle and again undo and redo the change to get a feel for which direction is improving your score.

Joined: 12/07/2007
Groups: Street Smarts
Play it again Sam

Why are you so committed to the model you have? You may have already squeezed, tweeked, and extracted all of the possible points from it. ...but it is possible that a more radically different model would do better. Look for hydrophobic (or -philic) sidechains that are out of place. These are clues to you that a better model exists. But you may not be able to reach it without some radical changes. So, if you find yourself not making any progress on points, save a copy of your model, and pull it apart and see if you can make another one that better fits things together.

Picture a mouse in a maze. He may be very very close to the exit, and have a very high score. But it is possible that the only way for him to actually get out is to backtrack and try another path. Picture your protein as a long string that has been folded onto itself. The order of the folds can dramatically change the fit. And in order to find out, you have to unfold it, swing things around and try another alignment.

Joined: 12/07/2007
Groups: None
Another Tweeking Tip

I find that as I get the model more and more compact, that using the drag tool then wiggling the backbone makes the score drop.  Some times locking the side chains using the lock all action before using the drag tool keeps the side chains from being pushed out of alignment by the drag and improves your score.  Try it the next time you get stuck and anything you do drops your score, it may help, or not.

xavier's picture
User offline. Last seen 7 years 49 weeks ago. Offline
Joined: 11/16/2007
Groups: None
fold it! tip

greens in, blues out, drag->shake all->wiggle backbone->rinse->repeat

Joined: 11/12/2007
Secrets of Wiggle Backbone

Hey folders,

Here's a little insight I got into the wiggle backbone tool. If you click wiggle BB and everything blows apart, don't worry! It just means that there's something bad with your protein and wiggle BB is overreacting trying to correct it. Here's the trick:

Repeatedly undo (control-z on windows or command-z on mac) to before the wiggle and redo (shift-control-z on windows or shift-command-z on mac) to after the wiggle until you can figure out exactly where your protien blew apart. That should tell you where your problem is. Fix that problem (moving the sidechain, or gently dragging the backbone), and then try wiggle BB again.

Repeat this process until wiggle BB is well behaved. :)

Joined: 11/16/2007
Groups: Window Group
Rubber Bands

When you have a loosely folded starting structure (i.e. no pre-formed beta sheets, a bunch of helices tangled together) I often find it helpful to set up a network of rubber bands and pull, hopefully "snapping" the protein into an approximately correct conformation, then do the usual wiggle & shake combo to get things packed down.  It worked out pretty well on puzzle 21 for getting the bizarrely-ordered beta strands into place.

 - Chris

feet1st's picture
User offline. Last seen 7 years 12 weeks ago. Offline
Joined: 01/25/2008
Groups: None
Change your perspective

If you've tried a number of things and can't seem to see a higher score, try changing your perspective. Zoom in, or out. Rotate the structure around and view it from the other side. Change your viewing options to show (or don't show) voids.

...now that you've changed your perspective, go back and undo/redo your last few moves and see what effects they had from this perspective. You will often find that there were side effects occuring that you weren't able to see previously.

feet1st's picture
User offline. Last seen 7 years 12 weeks ago. Offline
Joined: 01/25/2008
Groups: None
Do the opposite

If you keep pushing the backbone inward and the wiggle backbone keeps moving it back out, and bringing your score back to where you started... stop doing that! Try pulling it outward once. Then try pulling it up, then down. Each time doing the wiggle backbone to see where it puts it back and how it settles in to it's surroundings.

If you keep dragging the same portion of the backbone to make your moves... try grabbing the opposite side of one of the nearby residues. If you were pulling inward, try moving a nearby residue, and pushing instead. This sometimes gives you some rotation of the entire coil you are working with, and pulls it around in to a better position then was possible from the one that seemed so obvious to you that you were using it over and over.

If doing so happens to prove difficult to get the motion you were attempting... perhaps that is a clue that what you are picturing is not how the protein will behave.

feet1st's picture
User offline. Last seen 7 years 12 weeks ago. Offline
Joined: 01/25/2008
Groups: None
Go easy on the clutch!

When I was first learning to drive "a stick" (manual transmission vehicle), I would practice by driving down (and back up) our country driveway using ONLY the clutch, no gas pedal.

I think some people may be jumping into the Fold it! driver's seat and "flooring it". Try a gentler, slower touch. When you do so, you will see movements and reactions you did not expect, these are clues to some of the forces within your model.

feet1st's picture
User offline. Last seen 7 years 12 weeks ago. Offline
Joined: 01/25/2008
Groups: None
Full adrenalin RUSH

Lend me your ears for about 2 min. and I'll "show" you something that seems to help score higher with most any puzzle.

"Undo" back to your best score.

Write down your current best score.

Begin wiggle backbone.

leave it wiggling...

Now, drag a very short little band out from a piece of backbone (drag a right click) to the background and leave it for a moment. Direction doesn't matter so much. But make it almost as short as you can.

Now draw another band on the opposite end of the thing and pull the band in opposite direction of the first. Again, very short length.

Now stop wiggle, "clear bands" and wiggle until the score halts again.

If you can nail the x's and delete the bands as you go, you can keep wiggling for VERY long periods of time. It tracks your best score at any point during that period so any future "restore best" will find your intermediate high during the whole wiggle.

I call it "full adrenalin RUSH".

It's much like drag, drag, drag, wiggle... drag, drag, drag, wiggle... all in one ball of wax.

It basically causes the wiggle to fight with itself and explore hundreds of combinations in that whole area of the structure shape. ONE of those stands a good chance of being better then where you were before.

You will find cases where one little band doesn't cause anything to move... but your score drops. Then you clear it and score returns. This is a reflection of how the stress on that point of the protein was reducing the score. These invisible stresses are what you are improving upon (reducing) as your score improves.

You will find cases where it seems to wiggle higher and higher until it reaches your previous high (or very close), then it flips the whole thing around or shifts it over and drops the score down again... I call 'em "landslides". They seem to occur because you had a tight structure ("local minima") at that previous high. The protein doesn't distort from that point very much when pulled... just try the same process again from different angles at different points on the backbone.

Large proteins, such as CASP Practice 4-2 just need a little nudge, then clear the band and let it settle.

...now go back to drag, drag, wiggle. Lock, drag, drag, wiggle...shake. Then come back to adrenalin rush.

Wildcat's picture
User offline. Last seen 6 years 34 weeks ago. Offline
Joined: 05/16/2008
Groups: None
More Rubber Bands, Bending Sheets, Stretching Helices

Forgive me if these "hints" are either self-evident or "common knowledge". :)

It is not necessary to anchor both ends of a rubber band into two different objects -- they can be anhcored into two different sections of the same object, or one end can be anchored in empty space!

In a fit of experimentation with the Strep Binding puzzle, I anchored both ends of one of the flat sheets, then hit "wiggle", and watched for a few seconds, as the band drew the ends together, forming a "bow"! For less "severe" bowing, I would "wiggle" it for only a second or two then stop.

Again in Strep Binding, I chose one end of the entire structure, and anchored it as far to the left as my screen would allow, then did the same on the right, and hit "wiggle", and watched as it unfolded everything into a long, untangled strand.

Another experiment involved stretching out a helix by anchoring the ends to points further out in space, and successfully stretching it a little bit. Taken to the extreme, you can uncoil a helix beyond repair -- until you use "undo", of course. :)

Helices can also be bent, by anchoring both ends, and then pulling the middle.

None of these have proven to be "game-breakers" for me -- in fact, I've only managed to squeeze out a couple of extra points that I had otherwise struggled to score using one or two of these applications. However, in the right hands, there's no telling what can happen. :)

Wildcat

Joined: 11/12/2007
Cool video.

This is a great video, if you have any more instructional (or show off) videos using a more recent version of the game, I'd love to post them on the video page.

Joined: 11/12/2007
Call for Videos.

I've created a forum topic soliciting videos. Post links to your videos there!

Joined: 05/19/2008
Groups: UnScientists
One thing that seems to work

One thing that seems to work good for me:
Place a rubber band between two areas, usually at opposite ends of the structure to compress it together.
Wiggle backbone, but only for about a second... as soon as you see more than one clash is a good time to stop.
Clear the rubber band and wiggle backbone (first WITHOUT shaking sidechains!)
THEN, if that does not gain you any points, click Undo, and this time Shake sidechains first, then wiggle.

Actually, with most moves I'll first try wiggle alone, then if that yeilds no results: undo,shake,wiggle)

Sitemap

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, RosettaCommons