Scientists have developed a novel approach to computationally infer previously undetected behaviors within complex biological environments by analyzing live, time-lapsed images that show the positioning of embryonic cells in C. elegans, or roundworms. Their published methods could be used to reveal hidden biological activity.
Their process leverages deep learning techniques to study cell movements, guided by simple physics rules similar to video-game play. “We observed new features of an unknown migration mechanism, called sequential rosettes, that were validated by biomarker experiments,” said Dali Wang of Oak Ridge National Laboratory who led the research.
“We used hierarchical deep reinforcement learning and convolutional neural networks to study the movement of the nuclei and then investigated the migrating cell within the simulated biological environment to discover what’s unknown in the system,” he said.
ORNL, the Sloan Kettering Institute and the University of Tennessee will continue developing their deep learning method to better understand other biological unknowns.