(
A) Identities (top labels) predicted by our CRF_ID framework overlaid on the image (max-projection of image stack shown). Data comes from strain GT296. (
B) Cumulative variance captured by traditional principal components (PCs) and sparse PCs. Sparse PCs capture lower variance as a tradeoff for minimizing mixing of different temporal dynamics across components to improve the interpretability of each PC. (
C) The weights of cells across first three sparse principal components (SPCs). Blue and red bars in SPC2 and SPC3 denote cells with significant non-zero weights in SPC2 and SPC3. Activities of these cells are shown in
Figure 5E. (
D) Left panels top and bottom – Y and X displacement of randomly selected cells in the head ganglion, blue to red ordered from anterior to posterior. Right panels top and bottom show corresponding velocities along Y and X directions. Smooth displacement of cells and phase shift in peak velocities of cells along the AP axis show signatures of wave-propagation in partially immobilized worm in microfluidic device. (
E) Mutual information between worm velocity and lagged activities of cells grouped as SPC2 positive (blue) and SPC2 negative (red). Positive lag of neuron activities at which mutual information is maximum indicates neuron activities precede velocity. Shading indicates mean ± sem. The horizontal dotted line indicates entropy of velocity that is maximum mutual information any random variable can have with velocity.