Signal extraction from movies of honeybee brain activity by convex analysis

Calcium-imaging enables us to record movies of brain activity from the antennal lobe, a region of the honeybee brain responsible for processing odor information. Here, we present a matrix factorisation framework to automatically detect the neural units in this region and to accurately estimate their signals. Based on a non-negative mixture model, the algorithmic approach is to construct a convex cone that contains the data. The generating vectors of the cone are the purest, least mixed timeseries from the movie and serve as basis vectors for the matrix factorisation. We show that vectors selected in this way correspond to the biological signals and evaluate the method on both artificial and biological data.