Kyler Sood

Signal transduction regulates a diverse group of cellular functions, such as communication and internal processes. Signaling activity strongly varies with time, cell specialization and shape, and the exterior environment, and many drug mechanisms rely on inhibiting cellular signaling. We propose cellular harmonics, a graph-based signal processing method to quantitatively examine and classify cellular signaling, while being robust to the vast variations in shape within cell families. First, immature and mature dendritic cells were imaged using photoactivated localization microscopy (PALM), and the images were segmented and smoothed to isolate the cells from the background. Immature dendritic cells have filopodia, which are thin, spindle-shaped projections used in signaling, while mature cells use broader projections to signal. Pixels were uniformly sampled from the exterior of the cells and used to construct a graph, and the cellular harmonics were computed as the eigenvalues from the graph Fourier transform. The eigenvalues are interpreted as measuring the energies in the underlying structure of the cell, and the filopodia associate to a set of smaller energies. This method was found to reliably distinguish immature and mature dendritic cells and is invariant to rotation and scaling, and studies on synthetic graphs demonstrated broader applicability. Ultimately, cellular harmonics is a graph-based unsupervised learning method to distinguish cell classes and signaling activity, which can be applied to drug discovery. Future work will apply cellular harmonics to video captures of cells to examine transient cellular processes and images of fluorescent-tagged signaling molecules.