Jessica Fournier has a job that makes poor use of her talents. She spends her days stocking sneakers at a warehouse outside Grand Rapids, Michigan. A decade ago she was an astrophysics student at Michigan State University, where she coauthored a paper on RR Lyrae, a low-mass star that pulsates light. But having failed to secure long-term employment in her arcane field, today she pays her bills by cataloging shoe sizes.
She may have given up astrophysics, but Fournier still has a deep love of science. As soon as she gets home from work each night, she boots up her Asus laptop and begins what she calls “my second job”: designing molecules of ribonucleic acid—RNA—that have the power to build proteins or regulate genes. It is a job that she happens to perform better than almost anyone else on earth.
Under the fitting nickname “starryjess,” Fournier is the world’s second-ranked player of EteRNA, an online game with more than 38,000 registered users. Featuring an array of clickable candy-colored pieces, EteRNA looks a little like the popular gameBejeweled. But instead of combining jewel shapes in Tetris-like levels, EteRNA players manipulate nucleotides, the fundamental building blocks of RNA, to coax molecules into shapes specified by the game. Those shapes, which typically look like haphazardly mowed crop circles or jumbled chain-link necklaces, represent how RNA appears in nature while it goes about its work as one of life’s most essential ingredients. No self-sustaining organism gets made without the involvement of RNA.
Tweaking molecular models in this fashion is surprisingly fun—and, it turns out, useful. EteRNA was developed by scientists at Stanford and Carnegie Mellon universities, who use the designs created by players to decipher how real RNA works. The game is a direct descendant of Foldit—another science crowdsourcing tool disguised as entertainment—which gets players to help figure out the folding structures of proteins. EteRNA, though, goes much further than its predecessor.
The game’s elite players compete for a unique and wondrous prize: the chance to have RNA designs of their own making brought to life. Every two weeks, four to 16 player-designed molecules are picked to be synthesized in an RNA lab at Stanford. “It’s pretty incredible to imagine that somewhere there’s a piece of RNA that I designed that never existed anywhere in nature before,” says Robert Rogoyski, a New York City patent attorney who has had 14 of his EteRNA designs selected for synthesis. “It could encode a protein that no one has ever seen, something that’s important in the discovery of the next blockbuster glaucoma or cancer drug. Or it could be the cause of the zombie apocalypse.”
The chance to win this reward has proven highly motivating for EteRNA‘s players. They carefully study the data that the lab provides on how the synthesized molecules behave when ushered into existence, then use their observations to refine their next designs. In doing so, they—like their Foldit-playing peers—have helped scientists take advantage of the human brain’s unparalleled talent for recognizing patterns and solving puzzles. But EteRNA players have also done something much more profound: By scrutinizing their creations, learning from their triumphs and mistakes, and using their accumulated wisdom to develop new hypotheses, they aren’t just building better RNA molecules; they’re discovering fundamental aspects of biochemistry that no one—not even the world’s top RNA researchers—knew before. And in doing so, they are blurring the line that separates gamer from scientist.
At the heart of EteRNA is a small rivalry, the sort that often forms in competitive academic environments—in this case, inside the lab of David Baker, a renowned biochemist at the University of Washington. Baker has dedicated his career to figuring out and manipulating the byzantine shapes of proteins, the compounds that help form cells and make them function. Proteins consist of long chains of amino acids that fold into ornate spirals and loops as their constituent atoms push and pull on one another. Since a protein’s shape is vital to its function, researchers are constantly striving to grasp the rules that govern these contortions.
Adrien Treuille first came to Baker’s lab in January 2007 as a postdoc, having just completed a PhD in computer science. Treuille was drawn to the lab by Baker’s use of Rosetta@home, a free screensaver that doubles as a contribution to science. When a computer installed with Rosetta @home falls idle, the program can access the machine’s processing power to crunch data on protein shapes. By the time Treuille joined the lab, the screensaver had been downloaded more than 75,000 times.
There was just one problem with Rosetta@home: From the perspective of the person who installed it, it was totally inert. The screensaver did nothing but show an image of a folding protein. Treuille and a colleague named Seth Cooper were enlisted by their academic mentor, Zoran Popovíc, to add an interactive element to the program.
Within a few months, Treuille and the team had created a demo that allowed Rosetta@home users to earn points by manipulating a three-dimensional protein model into a specific shape. This interactive Rosetta was hopelessly buggy, prone to crashing twice an hour. But Treuille noticed that his testers were saving their shapes every minute—they were so absorbed in the quest for points that they didn’t want to lose their progress when the program inevitably conked out.
However, one postdoc in Baker’s lab, Rhiju Das, was a vocal critic of the game. “Rhiju was sort of this rock star, just this really smart guy,” recalls Treuille, a wiry 33-year-old New York City native who favors sagging jeans and hoodies. Das, the lab’s RNA expert, found the game pitiful. Because Das was held in such high esteem in Baker’s lab, Treuille was stung by his negative take on the new interactive elements of Rosetta. “Rhiju was a scathing critic,” Treuille says. “He basically wrote this email to the whole Baker lab where he was like, ‘I’m not sure how competent these computer science guys are, because this is a giant, steaming piece of crap.’”
But slowly the bugs were worked out, and Das was won over by the final version of the game, which was renamed Foldit prior to its release in May 2008. Tens of thousands of players flocked to Foldit that summer, drawn by the Rubik’s Cube-like challenge of twisting proteins until their most stable shape was achieved. By swapping tips, the best players quickly learned to beat the game’s most demanding challenges. Impressed, Baker urged these Foldit buffs to enter an international protein-folding competition, in which competitors—until that year always professional scientists—vie to accurately predict a protein’s shape based only on its amino acid sequence. By using the game to test the validity of their entries, a team of Foldit players outperformed many of the scientists—and scored first place in one category. Today the game continues to produce valuable data: In 2011, University of Washington researchers used insights gleaned from Foldit to crack the structure of a protein involved in the maturing of retroviruses like HIV.
In the fall of 2008, Treuille and Das went their separate ways: Treuille joined the computer science faculty at Carnegie Mellon, while Das founded his own RNA lab at Stanford. Despite their early quarrel over Foldit, the two young professors kept in touch; they would talk every few weeks to commiserate about the ups and downs of academic life. During one of these conversations, Treuille mentioned that one of his graduate students, Jeehyung Lee, was working on a Foldit hack that allowed players to manipulate RNA.
RNA was once dismissed as a sort of biological middleman, responsible only for carrying DNA’s instructions to the cellular machinery that synthesizes proteins. But now it’s clear that scientists had vastly underestimated the importance of RNA. Aside from the ferry-service type, there is also something called noncoding RNA, which among other things controls the expression of genes, amplifying or quieting their signals. Learning how this noncoding RNA works its magic will be a key to treating diseases caused by defective genes, as well as to understanding how life evolved.
Like proteins, noncoding RNA consists of long strands that warp into elaborate shapes which define the molecule’s function. Yet because RNA was considered relatively unimportant until recently, precious little is known about how these shapes form—far less than is known about the folding of proteins. This lack of knowledge is a major obstacle for researchers, particularly those working in drug development; if they can’t predict how a given strand of RNA will warp, they can’t synthesize novel molecules that will work as intended. There are some algorithms for predicting RNA shapes, but they are all flawed—no surprise, given that their human authors understand only a handful of the biochemical rules that govern the molecule.
Having witnessed all the good that had come from Foldit, Das hoped that Lee’s RNA game might help in the prediction of RNA shapes. In August 2009 he and Treuille traveled to a biocomputation conference in central Washington, where Lee gave a talk on his mod of Foldit. Afterward the three men gathered in a cabin on the edge of the Wenatchee National Forest to discuss the RNA version, which looked great but seemed too unwieldy to attract a mass following. Das wanted something slicker.
The trio brainstormed into the night, coming up with a slew of ideas for gamifying RNA. One proposal they kicked around was to have snippets of RNA fight each other to the death, in the style of a Japanese combat game called Senshuken; another was to create a first-person experience in which players navigated the world as RNA molecules of their own design.
The ideas were starting to get truly bizarre when something occurred to Das: Perhaps the solution wasn’t to mess with the Foldit formula too much but rather to alter the game’s incentives. “I said, ‘Let’s have it so players design something that we can make right now,’” Das says. “I wasn’t willing to move forward with this unless we could do synthesis.”
This was not an outlandish proposition. Unlike proteins, RNA molecules are easy to synthesize in a lab. The procedure begins with the creation of a made-to-order piece of DNA, which is then replicated millions of times using the technique called polymerase chain reaction. The resulting clip of DNA is then treated with an enzyme known as T7 RNA polymerase, which transcribes it into RNA. Das estimates that the whole process costs his lab around $100 per sample.
The game that Treuille and Lee designed, which they christened EteRNA for its January 2011 debut, has two tiers. The first is a series of “challenge puzzles,” in which players earn points by creating known RNA molecules. When the puzzle starts out, the RNA model is just a string of yellow disks that represent adenine—one of the four nucleotides that comprise RNA. A player must flip a certain number of those A disks to uracil (or U, colored blue), guanine (G, red), or cytosine (C, green) so that the structure morphs into the target shape—a two-dimensional form full of circular loops connected by straight “stem” sections, made from pairs of nucleotides. The task requires a working knowledge of how nucleotides attract and repel one another. The more nucleotides a puzzle has, or the more labyrinthine the shape the player is trying to mimic, the higher its point value.
Once players exceed 10,000 points—a milestone reached by about 7 percent of the game’s users—they are granted access to the lab section of the game. Here players are asked to mimic an all-new RNA shape—one that researchers would very much like to learn to make. Each player’s attempt has a shot at being synthesized at Stanford. The selection process is democratic, and each lab-qualified player is able to cast eight votes per week. Designers jockey for support by giving their creations snappy names (Strange Bird, Crimes Scene, Ends Justify Means) and writing descriptions of their trademark moves (“Added guanine boost to the tetraloop,” “This design includes a GU closing pair at 55-79″). Ultimately though, the voting players try to judge whether a given sequence is likely to form properly in the real world.
After the winners are synthesized, Das analyzes how well their structures mirror the game’s target shape. Then he grades each one on a scale from zero to 100. The highest-scoring molecules are posted inEteRNA‘s equivalent of the Hall of Fame, where they are scrutinized by scores of jealous also-rans. This is how the game’s newbies learn the secrets of RNA design and how its veterans reach epiphanies so profound they stun even EteRNA‘s creators.
To better understand EteRNA, I decide to give the game a try. After working my way through the tutorial, during which I pick the screen name Genghis Jones, I embark on a challenge puzzle—in this case the game shows me a type of noncoding RNA that is overexpressed in tumors. I am then given a straight chain of 90 nucleotides, each colored yellow to represent adenine, and instructed to change the nucleotides until they form the target shape—which in this case resembles a tadpole with a goiter on its tail.
The basics are pretty simple: Adenine bonds to uracil, uracil also bonds to guanine, and guanine also bonds to cytosine, with a GC pairing being by far the strongest. So I start to flip every other yellow to a blue, hoping to create adenine-uracil pairs that will form the structure’s long, rowlike stems. Once I have converted a dozen or so As to Us, the RNA begins to take form: I suddenly have a big loop that constitutes the head of the tadpole. But the nucleotide stems that are supposed to come together to form the tail remain separate. After some trial and error, I manage to seal those strings together by placing a powerful guanine-cytosine pair at the tip of the tail. Once I do that, EteRNA celebrates my victory with animated bubbles meant to represent the gurgling inside a test tube. It is my first taste of what is known in game parlance as juiciness, a moment of positive reinforcement that encourages the player.
It is a heady feeling, one that instantly reveals how this Flash-based game could be as addictive as the most beloved iPhone time wasters. “I don’t get a lot of sleep,” confesses Craig Lollin, a high school biology teacher in Piscataway, New Jersey, who ranks in EteRNA‘s top 20. “My girlfriend says that EteRNA is my other girlfriend—Renè9e, she calls it.”
Yet no player’s dedication can match that of Eli Fisker, the game’s top-ranked player and the unofficial mayor of the EteRNA community. A 35-year-old who lives in Aalborg, Denmark, Fisker has been fixating on patterns for as long as he can remember—”I fell in love with the patterns on my father’s shirts when I was a kid,” he says. After school, Fisker trained to become a librarian. But like Jessica Fournier, his chosen career never panned out, partly because he has what he describes as “something similar to Asperger’s.” He was sent instead into early retirement, a uniquely Scandinavian sort of luxury.
When Fisker first read about EteRNA on an Australian science blog, he thought it sounded like the kind of diversion he might enjoy, given his affinity for games involving rows and stacks of colored objects; he had recently purchased an iPod touch solely to play Bejeweled while on the go. Fisker wasn’t too familiar with RNA—his college biology class had barely mentioned the stuff—but the game doesn’t penalize scientific ignorance.
Within a week, Fisker had solved enough puzzles to cross the 10,000-point threshold. Then one of his very first submissions to the design competition was selected for synthesis at Stanford. But like most early winners, the molecule was a failure—it didn’t come close to folding into the right shape. “In the first rounds, the players were much worse than the existing algorithms,” Treuille says. When he synthesizes the players’ proposed sequences, Das also makes designs suggested by Vienna RNA or Nupack, two top algorithms for predicting RNA shapes. The software soundly trounced the humans in the early months of 2011, generating molecules that folded into reasonably accurate approximations of EteRNA‘s targets.
But the balance of power between man and machine shifted quickly. Fisker and the other EteRNApioneers pored over the data generated by the Das lab’s experiments, which includes the total energy of a molecule and diagrams displaying exactly where a sample folded incorrectly. The players took note of which base pairs didn’t stick in the real world and which clung together so tightly that they distorted the structure. They discussed their theories in EteRNA‘s chat room and wrote up lengthy strategy guides. And as the weeks went by, their results steadily improved. By May 2011, even humans with middling scores were regularly destroying the algorithms.
Given what they had learned from Foldit, Treuille and Das were not surprised that EteRNA‘s players were coming out on top. Computers are excellent at tasks like chess, in which all the parameters are known—64 spaces on the board, pieces with finite moves (albeit a lot of them). But they flail at endeavors for which the majority of rules remain a mystery and a bit of guessing must take place. (As Treuille puts it, “Computers don’t have flashes of insight.”) The human brain, by contrast, can use intuition to feel its way around problems through trial and error.
What did catch EteRNA‘s creators off guard, though, was the players’ knack for identifying and defining the rules of RNA behavior. It is one thing for amateurs to help solve a tricky puzzle; quite another for them to discern important laws of nature—the kind of thing that could conceivably earn a student a PhD—then share their discoveries with the world.
After several weeks of studying EteRNA‘s winning designs, for example, Fisker noticed that the most successful molecules were those that placed guanine-cytosine pairs at the junctions where multiple straight stems of nucleotides feed into massive loops. But simply aping this strategy in his designs didn’t always work as he envisioned. Fisker was missing something.
And then it dawned on him: Orientation meant everything. Unless the guanine- cytosine pairs all faced in the same direction, so that red always alternated with green as you looked around the loop, the molecule wouldn’t form correctly when synthesized. This breakthrough soon turned Fisker’s designs into world-beaters—and, more important, enhanced biochemistry’s grasp of RNA. It was a fundamental rule that no scientist had ever discovered before.
Several other EteRNA champions have also uncovered major properties of RNA. Jessica Fournier, for example, was among a handful of players to discover the importance of variation in a certain type of molecule. “If the design has a lot of arms that are the same length, try not to make them too alike, or they could mispair,” she explains. This is because nucleotides continue to attract and repel one another even when they are separated by relatively large distances. Making sure those rows have little in common—that the nucleotide sequence of each is distinct—minimizes the risk of having a base pair pulled out of alignment.
Das didn’t anticipate this kind of discovery. “The pattern recognition stuff, OK, I knew people were going to be good at that from the experience we had with Foldit,” he says. “What I didn’t know was how they were going to interpret and use the data. But if you look at what they’re doing, it’s much better work than some of the best graduate-level scientists. What they’re doing, it’s really beautiful.”
Did I mention that EteRNA is addictive? After playing the game for a while, I found that I wanted to do little else with my spare time. I started to play for an hour every night after putting my son to bed, choosing to fiddle with Chlamydomonas reinhardtii RNA rather than have a beer with my wife. Then one night I stayed up far too late agonizing over recalcitrant loops in Arabidopsis thaliana RNA. I started charting my progress on Post-its affixed to my office wall, keeping track of all my little eureka moments&mdash”Eliminate more sticky GC pairs,” or “G one space over from neck, stabilize loop.” Genghis Jones shot up the global leaderboard.
I earned my 10,000th point during a Sunday-night session fueled by Jim Beam and peanuts. My wife caught me doing a pitiful version of the Dougie to celebrate.
With this I began to build never before seen molecules that would have a shot of being synthesized at Stanford. My initial efforts were embarrassments. One player derided my second attempt as a “Christmas tree,” EteRNA slang for molecules that feature too many guanine-cytosine pairs and are thus predominantly red and green. The teasing hurt, but it also taught me to keep an even distribution of pair types.
Finally, after weeks of failure and frustration, I received the news: One of my molecules, entitled “The Revolution of the Mobile Archer,” had taken seventh place in the latest round of voting. Das’ lab was going to synthesize my design.
It was a thrilling moment, but there is a downside to consider: Every such triumph stands to make the game much, much harder. That’s because EteRNA isn’t just a game; it’s also a distributed wetware-computing operation. The data from every synthesized design is collected by EteRNA‘s servers and will eventually be incorporated into new shape-predicting algorithms. Humans aren’t just competing against the machine—they also serve it as intelligence-gathering slaves.
Already, an inkling of that automated future can be glimpsed in an algorithm that Jeehyung Lee devised from a very elementary form of EteRNA data: the strategy guides that players write to keep track of their discoveries. Lee centralized and mined these guides for the best tricks, including Fisker’s stratagem for placing guanine-cytosine pairs at the intersections of stems and loops. Lee then coded those tricks into an algorithm called EteRNAbot. This formula has proven more accurate than either Vienna RNA or Nupack, suggesting designs that fold nearly as well as—and sometimes better than—those generated by the game’s most accomplished players. “I feel like our days are numbered,” says Robert Rogoyski, the patent attorney from New York. “There’s a complex sort of pattern-recognition process that you develop after playing 15 or 20 times, and it’s hard to reduce that into words. But I think we can reduce it into words. And if we can reduce it into words, we can make it math.”
The only question, then, is how long it will take EteRNA to produce what Rogoyski dubs “the Terminator algorithm”—a program better than any human. The answer depends on how quickly the game can amass data from synthesized designs. If EteRNA were to maintain its current pace, with Das making about 32 molecules a month, it would likely take decades to compile the requisite intelligence. But the game’s experimental tempo is about to pick up.
In the earliest days of EteRNA, after growing frustrated at missing out on Stanford a few too many times, several players began to discuss the possibility of synthesizing designs themselves. One of this group’s more technically inclined members even priced the necessary equipment, such as a thermocycler to carry out polymerase chain reactions. The cost proved prohibitive, but home-brewing molecules is quickly becoming a realistic option.
Sequencing a genome costs less than 1 percent of what it did in 2002, a plunge that reflects the astonishing affordability of tools and technologies that barely existed a decade ago. Sooner or later, industrious EteRNA players are bound to invest the time and money to make Das superfluous, by building improvised labs where they can make molecules as they please. Instead of competing for slots at Stanford, they will use the game as a simulator to determine which of their designs are promising enough to synthesize in the garage.
Far from dreading his impending marginalization, Das is invigorated by the prospect of EteRNA players running their own experiments. “Any way we can get them more resources to test their hypotheses is going to be really revolutionary for this field and for science in general,” he says.
In the meantime, Das is trying to provide some of those experimental resources himself. He is planning to scale up the number of player molecules he can synthesize at a time, from eight to 20,000. Using part of the $1.7 million in funding that EteRNA recently received from Google and the W. M. Keck Foundation, Das plans to purchase DNA printed onto glass slides, which will let him analyze tens of thousands of RNA molecules at a time in parallel.
But Das also points out that garage labs aren’t the only way for players to make molecules—he envisions them eventually enlisting the services of satellite labs. “You could give EteRNA users remote access to a workstation. It doesn’t have to be at Stanford—it could be in Siberia,” Das says. “And they press a button that’s like, ‘I want to make this piece of RNA,’ and it gets uploaded to the biochemical cloud and synthesized.”
By synthesizing molecules of their own design, EteRNA players will be doing work that was once the sole domain of scientists. With all this coming together, Treuille and Das figure, why not also implement the final piece of the scientific process: publishing. Das is in talks with the Public Library of Science to create a way for players to contribute papers to the online journal PLoS Currents using the data generated by the RNA synthesis. “Imagine on the website you start typing in the intro paragraph, and then the other paragraphs, and then you hit Create and Submit Paper,” says Treuille, who is now in the midst of a yearlong sabbatical at Google’s advanced-projects lab, Google X. “And it assembles the whole thing—puts in your data collected from EteRNA. And then it goes to the journal, and they know the data is true because they know that EteRNA is true.”
Once that happens, biochemistry will face a dilemma that other, far less technically demanding fields have been grappling with for years: Who exactly are the professionals? The line between computer-game player and scientist will be nearly gone. If someone can form a hypothesis, test it herself, and then publish her groundbreaking results in a respected journal, she is a scientist—even if she makes her living stocking shoes.
My molecule, “The Revolution of the Mobile Archer,” fared much better than I had hoped. Das scored it a 91 out of 100—higher than entries from such elite players as Eli Fisker and Jessica Fournier, not to mention those suggested by the Vienna RNA and Nupack algorithms. I came in five points behind the winning design—one suggested not by a human but rather by the EteRNAbot algorithm, which registered a rare victory in the competition.
When the diagram of my molecule in real life was posted on EteRNA, however, my joy turned to disappointment. The target shape resembled a two-headed stick figure lying on its side. But the lab diagram showing what I had actually produced looked more like the cartoon character Ziggy doing the splits.
I stared at the molecule’s most problematic region—the 16-nucleotide loop that formed Ziggy’s torso. A guanine adjacent to the loop was attracting a uracil from quite a distance, thereby creating a kink of nucleotides that shouldn’t exist. I wondered if I could prevent that from happening by changing that guanine to an adenine. Or maybe I just needed to flip the direction the guanine was facing, keeping in mind Fisker’s axiom about orientation.
The more I focused on that clump of nucleotides, the more I fantasized about a future in science. So what if I had struggled to dissect a fetal pig in high school biology class? I was on the verge of making an important discovery about the placement of guanines in RNA molecules, based on data from a real-world experiment. Who’s to say I’m not a biochemist?
Source: Wired
