Fig. S8

Clustering of local pixel correlations reveals highly correlated activity patterns for cells that were classified as multiple ROIs. (A) Examples demonstrating the effectiveness of the method. Four ROIs that were manually identified to be single cells (their activity traces are shown on the right) were combined and each pixel of these ROIs was correlated with each other. Subsequently, we performed principal component analysis (PCA) and expectation maximization clustering, which automatically segmented even spatially close neurons as independent units. Pixels were plotted according to the first two principal components (PCs) at the bottom, illustrating that that the developed algorithm (PCA and clustering) successfully identified the correspondence of the pixels to their original ROIs (lower left: color code based on manual ROI selection, lower right: color code based on assigned cluster identity). (B) Example cluster analysis showing a potential “multiple-cell” ROI (see manual analysis in Additional file 7: Figure S7) that was split into three separate clusters by the algorithm (left); and the accompanying activity traces for each cluster (shown on the right). (C) Example cluster analysis showing an ROI that wasn’t split, and its corresponding activity trace. (D) Quantification of the average correlation of the mean ROI traces resulting from the clustering (e.g., average correlation of the three traces shown in (B)). Correlation is overall high, suggesting that there is no major signal contamination even for neurons that were manually assigned to potentially contain multiple-cell activity. (JPG 2823 kb)