Fig. 1. Behavior, data collection and encoding model construction.

a Top: subject capturing prey in the apparatus. Behavioral video data was collected by five (MG) or two (TY) cameras. Bottom: subject behaving freely and spontaneously in the home enclosure. b Top: A raster plot shows the activity of 175 units recorded from TY, from one second before to 3 s after the duration of a reach (reach 3). Units are ordered as in (c). The green and blue bars above the raster indicate times corresponding to spontaneous behavior (green) or prey capture (blue). The gradient bars highlight periods of increased firing rates across the population. Bottom: hand position for reach 3 (black), with reach onset indicated by the black dot and the same gradient indicating neural modulation imposed on the kinematics. c A sample of hand trajectories for six of 101 reaches, separated into reachSet1 (dark blue) and reachSet2 (light blue). d FNs were computed from activity during either reaches in the corresponding reachSet or during spontaneous behavior. Within motor or sensory areas, units are ordered by the average in-weight to the target unit for reachFN1/2. The color scale corresponds to $w_{ji}=conMI$. Lower Right: Functional weights versus inter-electrode distance (mean ± sem), with sample numbers ranging from 4 samples at 4327 µm to 1176 samples at 894 µm (individual sample numbers are available in the Source Data file). e A representative sample for the trajectory and full kinematics models with ${\tau }_{lead}=100\mathrm{ms}$ and ${\tau }_{lag}=300\mathrm{ms}$. Left: wrist position for a single reach (black) overlaid with the spike sample time (white circle) and corresponding trajectory sample including lead (blue) and lag (red) movements subsampled at 40 Hz. The black dot indicates the average position for the trajectory sample. Right: velocity samples with the spike sample window and trajectory center shown with a vertical black dashed or dotted line, respectively. f The corresponding sample for the velocity and short kinematics models with ${\tau }_{lag1}=100\mathrm{ms}$ and ${\tau }_{lag2}=150\mathrm{ms}$. Source data are provided as a Source Data file.