Hold It! Results
Thanks for surviving my various biotin puzzles! I know some of the restrictions have made them very difficult. We've been experimenting with various puzzle setups and I think we'll soon find a happy intermediate between impossible and hard. :)
As for the results of the previous puzzles, I was pleasantly surprised with some of the solutions that were turned in. I'm curious to know whether any of you were able to find any strategies that seemed to work better than randomly searching. In future puzzles, I will include those solutions along with some others we have generated as the starting structures, with the primary goal being to maintain the hydrogen bonds while also pursuing the best packing score. We have a few more tweaks to apply, but be confident that I won't ask you all to do the impossible anymore. (find those bonds from scratch).( Posted by austinday 93 2664 | Sat, 03/20/2010 - 21:07 | 0 comments )
New Solution Voting Puzzles
CASP 9 is going to start on May 3rd, and we are very excited that this year Foldit will be its own CASP structure prediction group. Unlike CASP 8 where Foldit submissions were part of the Baker Team, this summer all of you will be part of 'The Foldit Team' competing against structure prediction labs from around the world.
As you might already know, CASP lets you submit 5 models for each target, so we have been trying to think of cool ways to let YOU, the Foldit players, pick these models (other than the top scoring Foldit solution which we plan to submit as one of the 5 submissions).
Today we are trying a new puzzle type where you get to vote for what you believe is the best model. In these "Solution Voting" puzzles, you will be given different high-scoring Foldit solutions from a previous puzzle. In today's case you have 5 Foldit submissions from puzzle 266: Mini-CASP 17.
You can look at each solution in gallery mode, rotating them around and comparing one solution to another using the guide, and then vote for the model you believe is better. You can show clashes, voids, exposed residues, bonds, and sidechains, but there is no score information given so you can't just vote for the one with the best Foldit score. All of these models are already very good-scoring Foldit solutions, and we're interested to see if you can pick out good models based on your own ideas of what proteins should look like.
Once the native structures are released, we will let you know which model was the closest to the native and will assign you points based on how close your vote was to the native. We haven't worked out all the details for that part yet, but the more "Solution Voting" puzzles you participate in and the more often you
vote correctly, the higher your Foldit rating will be. The skills necessary to rank good models might be a Foldit skill that we haven't measured before. As such, we might have to come up with a new ranking category:
We will test this out before CASP 9 begins, and if it seems to work well, Foldit might be able to participate in one of the other CASP categories: Quality assessment of models (http://www.predictioncenter.org/casp9/index.cgi#scope), where the CASP organizers give us a bunch of predictions and we have to rank them (which do we think is closest to the native). This could also be a very exciting category for Foldit players as well!
For now, please try out the new 266: Mini-CASP 17 - Solution Voting puzzle. We picked 5 different Foldit predictions from puzzle 266 for you to look at. For this first puzzle we expect to get the answer very soon, perhaps even this week. We don't know what the native looks like either, but since you've been working very hard on Mini-CASP 17 we hope you all might be able to make an educated guess!( Posted by beta_helix 93 2664 | Wed, 03/17/2010 - 05:54 | 0 comments )
Upcoming ligand binding puzzle!
We have an upcoming Foldit puzzle where we are again attempting to bind a biotin molecule! I hope you all didn't miss it too much! In this puzzle, we have introduced indicators to help the players make more ideal hydrogen bonding angles. For this specific ligand, we're asking for 3 hydrogen bonds to be made with the oxygen at the "business end" of the ligand. (The side that is directed into the scaffold). Here is an example of what the near ideal angles should look like:
We aren't so much interested with hydrogen bonds being made to the linker portions of the ligand. (Everything that is not at the business end of the ligand) But the more the better!
I hope you guys can come up with some interesting solutions! If you can make a good enough solution, we'll make the protein and tell you if it works! (Should take about a month from time of order) And if it works, we'll put you on the resulting paper!( Posted by austinday 93 2664 | Fri, 03/05/2010 - 23:03 | 2 comments )
First alignment puzzle results!
We've been looking at the results from the first alignment puzzle, and they are definitely encouraging! In this case, we know the native (the protein structure we are trying to find). Here's an image with the template from the puzzle in red, the native in green, and the top scoring structure (produced by TheGUmmer and Madde) in blue. So the question is, can we get from threading the red template to the green native?
You can see that the main difference between the red and green structures is that loop on the left side. But the blue solution structure has gotten much closer to the native! The alignment that was used to make the solution was:
For comparison, The starting alignment for the puzzle was:
So they're pretty similar, but sometimes small alignment changes can make a difference. Excellent work!( Posted by Seth Cooper 93 2664 | Sun, 02/28/2010 - 22:29 | 0 comments )
New Comparative Modeling Tools for Structure Prediction
This week we have released some new tools in Foldit for structure prediction. The update comprises several tools and some new levels that teach you how to use these tools. The rest of this blog post is to try and give an overview of why these tools are useful in structure prediction.
Often when we try to predict the structure of a protein, that protein has a similar amino acid sequence to one or more proteins of known structure. These known structures are called templates, as they can be used as starting points for predicting the structure of new proteins. We call this approach to structure prediction comparative modeling, as it is based on comparing a protein sequence to existing structures in order to build models. These tools are based on the same code that is being used within the Baker Lab for comparative modeling and structure refinement, so that comparisons between automated Rosetta protocols and manual structure prediction using Foldit are more fair now.
We compare protein sequences using a tool called a sequence alignment. A sequence alignment is simply a mapping of residues from one sequence to another. For comparative modeling the sequence alignment tells you which residues are analogous between the query sequence and the template sequence. In the context of comparative modeling, the sequence alignment tells you which sections you should copy from a template structure and which residues you should try to optimize more freely. By manipulating the sequence alignment, you can get different starting structures which you can then refine using the existing tools in Foldit.
Many of the successful predictions from CASP8 were comparative modeling puzzles, but sequence alignment for these targets were done by scientists in the Baker Lab and left fixed for the Foldit players. With the latest Foldit tools for comparative modeling, you can manipulate the sequence alignment and use it to generate models interactively. Our hope in releasing these tools is that by being able to explicitly manipulate the sequence alignment and refine the resulting models, the players can both understand structure prediction better and build good comparative models of proteins using Foldit.( Posted by tex 93 2664 | Tue, 02/23/2010 - 20:01 | 0 comments )