Gamers outshine computers in crowdsourced RNA modeling

Results from the crowdsourced game and experiment, EteRNA, which combines RNA folding puzzles with laboratory synthesis, show that human gamers are able to develop better models of RNA folding than previous computer algorithms. Design rules formulated by the online community have even been used to construct a new algorithm, EteRNABot, and in some cases represent completely new understandings about RNA folding that have yet to be explained mechanically.

Three years ago, Carnegie Mellon University and Stanford University created EteRNA as a new way to empirically develop designs for synthetic RNA that would fold into certain shapes. RNA is similar to DNA and understanding its functioning and folding is critical for research into HIV, retroviruses, antibiotics, cancer, and heart disease.

While crowdsourcing simulation data is common and as easy as having participants run programs on their own computers, such as with Seti@Home, EteRNA goes an extra step, getting participants to personally solve folding puzzles as games instead of relying on their computer’s processing power. In fact, the human touch is what seems to make the project successful.

Unlike an earlier protein-folding game, FoldIt, community designs in EteRNA are synthesized in a biochemistry lab at Stanford to physically test the solutions and give additional feedback to the players.

Not only were the community’s designs 99 percent likely to be better than those created using existing RNA folding algorithms, but the community developed novel hypotheses about folding rules.

A researcher on the project, Jeehyung Lee, said the quality of the designs was so high that "even if you generated thousands of designs with computer algorithms, you'd never find one as good as the community's." However, a solution that might take a player a day or two to develop could be solved less perfectly in a minute by an algorithm.

With the realization that the community was obtaining information not yet compiled into these existing algorithms, the team created the EteRNABot algorithm which utilizes the community's design rules. As an example, a common rule "broken" by the previous algorithms was, essentially, "close all multiloops with a pairing of guanine and cytosine. The first closing pair should put cytosine on one side and in additional pairs, this should switch."

Even after doping the EternaBot AI with these rules, the humans still outperformed the bot they helped to create. It is interesting to note that the participants are described primarily as not specifically trained in biology and collectively underwent a learning curve before obtaining their current mastery.

EteRNA is next going to have puzzles that incorporate 3D folding added, along with a template to allow researchers at other organizations to submit puzzles of their own research projects.

It’s not too late to start playing EteRNA, which is a flash-based game hosted on CMU’s servers.

The paper of this research was published today in the Proceedings of the National Academy of Sciences.

In the video below Adrian Treuille at CMU discusses the purpose of EteRNA.

Sources: Carnegie Mellon University, EteRNA