Diversifying Protein Structure Prediction in Foldit, a Human Computation Game

Foldit is a puzzle-based human computation game, built on top of the Rosetta molecular modeling software suite, that explores the application of human problem-solving skills to protein structure prediction. In Foldit, players make anywhere from a few dozen to upwards of ten-thousand moves in order to solve protein conformation puzzles, the results of which are aggregated by researchers and used to predict the three-dimensional structure found in nature, otherwise known as the native conformations. The purpose of this research is to determine if limiting the number of moves a player can make over the course of a puzzle, and thereby limiting the depth of their search, the player can be encouraged to create more diverse, numerous, and/or accurate predictions. In order to test this hypothesis, a new feature was implemented to track the number of moves players make while solving a puzzle, and penalize solutions whose cumulative moves surpass a set limit. Side by side comparisons were made between the results of experimental (group) puzzles that included a 250 move limit and control puzzles that did not include any limit. Preliminary results from this ongoing study suggest that additional data will be necessary to determine the exact impact of the move-limited puzzles on structural diversity and prediction accuracy.