Fig. 7: Divergence between neural activity subspaces aligned with inactivation effects, muscle activity and limb kinematics.

a, Schematic of different scenarios for the overlap between neural activity subspaces. Dim, dimension. b,c, Canonical variables for muscle activity (b) and neural activity (c) for one mouse. Max., maximum; min., minimum. d,e, Canonical variables for limb kinematics (d) and neural activity (e) for the mouse used in b and c. f, Correlation coefficient (black) and fractional muscle activity variance captured (red) for canonical variables. Each set of connected dots in f–n is from one mouse. g, Cumulative fraction of muscle activity variance captured by canonical variables after orthogonalizing their corresponding vectors. h, Correlation coefficient (black) and fractional limb kinematics variance captured (red) for canonical variables. i, Cumulative fraction of limb kinematics variance captured by canonical variables after orthogonalizing their corresponding vectors. j, Overlap of different activity subspaces (black circles) compared to 100 estimates of the overlap expected by chance for each animal (gray dots). k–m, Principal angles for different activity subspaces (black circles) compared to 100 estimates of the principal angles expected by chance for each animal (gray dots). k, Effect versus muscle. l, Effect versus limb. m, Muscle versus limb. n, Mean overlap (black circles) between subspaces defined from two maps made with separate halves of time series segments, for 300 different segment parcellations (gray dots).