Reconstruction of raw neural activity in Thy1-GCaMP6f mouse using Nonnegative least squares decomposition. Left: Raw GCaMP activity. Middle: Spatiotemporally unmixed linear model, created by multiplying temporal H (obtained by k-means clustering into 18 ROIs) with spatial W (obtained as an output of nonnegative least squares, where H was used as the input). Right: Residual, obtained by subtracting the raw from the reconstruction.