Report from the first Dev Meeting 2012 with players

Started by auntdeen

auntdeen Lv 1

A week ago the devs had the first meeting that they invited players to Skype into, and I was fortunate enough to attend. Apologies for lateness (been a very busy week in RL), but I would like to let the community know what was discussed with myself and Coyotee, the other player present.

First, beta_helix showed us a graph that had the distribution of all foldit players for the first short alignment puzzle (please see the attached pngs below). It clearly shows that the players did better than rosetta.

He also said that the alignment native (which was recently released) had 3 disulfide bonds… that the best scoring foldit solution had two, which is excellent, and many of the top scoring solutions had all three disulfides!

And, that the disulfides DO score well now.

There was discussion about the new alignment tool and the cutpoints. The devs are hoping that some in the community can get expert at accomplishing them soon (especially with partial threading) and hopefully do a video so that the rest of the players can increase their understanding of and ability to use these tools well.

We briefly discussed what the best place would be, online, to post these videos in an HD format, whether on the foldit website or uploaded to a video host (youtube, vimeo, etc).

The devs spoke of hoping to use webinars for advanced techniques with the webinars conducted by expert players. The alignments/partial alignments would be a priority for these.

That led to a discussion of meetups. My team has done a few regional meetups, and I had offered before and offer now to help set up "All Foldit Players" meetups in regions of the world where we have enough concentration of players to make that possible. The devs expressed that where & when possible, at least one would try to attend large meetups for players.

They would also like, at some point in the not so distant future, to try to have a small group of players go to Seattle to meet with all the devs. They mentioned that they might do something like choose from the top of the leaderboards.

My response to that was to caution that we still need to have diversity score better than it currently does - which they realize and agree, and are working on that.

So then - here are the questions for the community:

  1. Is there anyone who has mastered the alignment tool - and partial threading - who is willing to do a video and/or a webinar for the global community?

  2. I know that some people had started maps of "where-in-the-world" players reside… Is there anyone who could coordinate an overall map for use to determine good global (All Foldit Players) meetup locations? Or is that something better left to the devs to develop?

  3. Are you interested in attending a meetup of all foldit players in your area? What distance are you willing to travel?

  4. Are you interested in coordinating a meetup?

Please post any comments/requests for clarification here :-)

Susume Lv 1

Could we get an explanation of what is on each axis of the graphs? Auntdeen tells me the lower right corner of the first graph means the best solutions, and the lower left corner of the second graph is best.

beta_helix Staff Lv 1

In both graphs the y-axis is the Rosetta energy (lower is better and it is the same energy function used in Foldit).

Both graphs have a different metric on the x-axis to describe "how close to the native solution" each prediction is. On the left is GDT (http://en.wikipedia.org/wiki/Global_distance_test) and on the right is the RMSD (http://en.wikipedia.org/wiki/Root_mean_square_deviation_(bioinformatics)) over just the CA-atoms.

So in both plots you want to be as close to the middle of the screenshot as possible (the best GDT score would be 1.0 and the best RMSD would be 0).

As you know, when folding you have no idea what the GDT or RMSD of your model is, the only thing you are guided by is the score… this is why IDEALLY the lowest points in both graphs are also the closest ones to the native (since those are the models that we can easily pick out: such as the top-scoring Foldit prediction).
It is rarely the case that this works out (the best-scoring model being the closest to the native) and you can see that the lowest Rosetta energy prediction (the lowest purple dot on both graphs) is NOT the closest to the final solved structure, but even that top-scoring model is closer to the native than all the Rosetta predictions (which is great!). And many of the other top-scoring Foldit predictions are even closer to the native!

spmm Lv 1

It would be fairly trivial to manually put the City, State, Country location on a map for the 300 or so regular players, assuming that players had provided that information on their player pages or were prepared to provide it for the map. Were you thinking of having a player nic attached to each pinned location or just a cluster style map? Would be quite sparse if only the regular players are included.

Not sure if you were talking about mapping the thousands of non regular players since 2008, that would possibly be best done anonymously using IP addresses, if they have been logged somewhere, and would probably be inaccurate much below the country level and therefore not a privacy concern. Web servers usually log some location information.

As always on the public interweb people should be careful about providing detailed personal information like their residential address to be used for geocoding a mapping app or mashup.

jeff101 Lv 1

You could propose a number of sites around the world for having meetings and then let folks vote for which ones they would most like to visit for a meeting.

beta_helix Staff Lv 1

…but we started off that meeting with a word from the Scientists in the Baker Lab working on the Flu.
This was their message to all of you:

Thanks to all the Foldit players who worked on the recent Influenza puzzles!

There were quite a few interesting results among the solutions proposed, and we're in the process of computational fitting these solutions onto existing scaffolds. We hope that in a few weeks we'll have several hundred computational designs based on these Foldit solutions, which we will order and experimentally characterize in the coming months. We're working hard down this path and will keep you updated on our progress!