The algorithm and data analysis pipeline. (A) Demonstration of the ARCOS algorithm on a growing activity cluster. Protein activity of cells arranged in a mesh can assume two states, inactive (light gray) and active (black). In step 1, the dbscan algorithm spatially clusters active cells independently in each frame. In step 2, clusters are tracked between frames. The cluster in frame 1 forms a seed of a CE. The cluster in frame 2 is linked to this seed cluster because its member cells are within the neighborhood radius εprev. In frame 3, only clusters #3, #5, and #6 are linked to the previous frame's cluster. Cells in cluster #4 are too far and thus form a new seed of a CE. (B) Flowchart of the core algorithm for detecting CEs in time-lapse data. (C) Image acquisition and analysis pipeline. Segmentation and tracking software identifies cells in multichannel time-lapse movies and produces a long-format output where every row is the measurement m in a cell ID at coordinates (x,y), at time t. Single-cell protein activity time series are then detrended and binarized. Binary activity time series are then fed into ARCOS to identify collective activity. (D) Detrending and binarization of single-cell ERK activity with a running median method. Time series are first smoothed with a short-range filter, then the signal smoothed with a long-range filter is subtracted from the original. The result is rescaled and thresholded to binarize it. Red segments indicate periods of activity, which are then processed by ARCOS. (E) Binarization of a single frame with a LISA, G* statistic. The original frame is compared to the same frame with the measurement randomized between cells while keeping the original X-Y positions. Black dots indicate active cells with G* > 2.