Video game helps researchers determine protein structures
Many people spend countless hours and large amounts of energy playing video games. To most, video games are fun and relaxing, although many others see them as unproductive. But what if there were a video game that was not only fun and competitive, but helped in solving complex and important problems to benefit health and science? Two Carnegie Mellon researchers have designed a video game to do just that.
Adrien Treiulle, a faculty member in Carnegie Mellon’s department of computer science, and one of his Ph.D. students, Jeehyung Lee, have developed a video game to help solve one of the most complicated problems in biochemistry: Protein structure.
“A protein is one of the most fundamental molecules in our body. In order to figure out how the protein interacts and how it works in our bodies with other molecules, it’s important to know what it actually looks like,” Lee said. By nature, he explained, a protein always folds into a shape that minimizes its free energy. This is its most stable conformation.
The game, conveniently named “Foldit,” has players manipulate protein structures to reach these optimal, lowest-energy structures. The better the structure they produce, the better their score, which is always displayed on the screen. The player’s score is compared against other players’ scores in real time, instilling a sense of competition among all of the gamers.
The benefits of understanding the structure of proteins are numerous and extremely important. “There’s both the scientific goal of understanding how these proteins fold, and also the engineering goal,” Treiulle explained. “If we can reverse-engineer these principles, then we can modify proteins and create new ones…. We can really engineer cellular mechanics.” By understanding the structure of proteins, new proteins could be created or modified to help the sick or even cure certain diseases.
But why are humans so important in this process? Computers are robust, consistent, and persistent; it would seem as though they possess every skill needed to solve complex problems like protein structures. However, computers lack one important quality: intuition.
When solving for good protein structures, computers essentially scan over every part of the protein and give each part of it equal attention, trying to find the best solution to the problem at that specific location. Humans, however, can focus on “problem spots,” ignoring the parts that don’t need as much attention.
“To a human, pulling that one [problem spot] out is one intuitive step. But to computers, that step is really far away. Humans can make more intuitive, creative decisions on the global level, not just the local level. This makes a huge difference for computers. The problem is computers always try to find the locally optimum solution,” Lee said. “Sometimes computers can never find the [correct structure]. It just gets stuck on some local optimal solution.”
Treiulle and Lee recently co-authored a paper that was published in Nature that discussed the results of Foldit. It was found that the players used a wide range of methods to perfect their protein structures, which varied within individual proteins as well as across different protein puzzles. This differed from the way the computer searched for an optimal structure, and as a result, players were generally able to find better protein structures than the computer could alone.
Interestingly, Treiulle and Lee both admitted that the vast majority of Foldit players have no educational background in biochemistry at all. In addition, in a post-study survey, some players stated that their contribution to science was a motivating factor to play the game, but many other players took notice mainly for the competition.
Although there are some improvements to be made to this human-computer interactive technique, such as instilling more collaboration among the players, it is the first of its kind and shows quite a bit of potential for the future of computing. “We have this three-way meld of human creativity, computer science, and machine learning — all braided together and forming this new way of solving this problem, which has never been done before,” Treuille said.