Figure 9. Hand velocity decoding from neural activity in motor cortex and deep cerebellar nuclei (DCN).

(A) Observed (left column) vs. decoded (center cortex, right DCN) velocity trajectories in each direction for a session with simultaneous recordings. Blue lines represent all trials with pontine nuclei (PN) perturbation (‘laser’) and their mean (thicker line). Orange lines represent all ‘test control’ trials, not used for training the decoder, and their mean. (B) Observed vs. decoded mean differences in velocity between laser and control test trials, for each of the sessions with cortical recordings (n = 38). In the heatmaps (left observed, center decoded) rows are sessions, ordered based on the observed mean velocity difference in the forward direction in the 100 ms following lift. Orange regions indicate time points in which the velocity is higher on control trials, and blue regions indicate points in which the velocity is higher on laser trials. The row denoted with a mark is the session shown in (A). Right panels show the distribution of rank correlations (Spearman’s ρ) between the observed and decoded differences in velocity over time (i.e., between corresponding rows of the heatmaps). Within each panel, individual correlations and their median (black line) are shown on top, with p-value of the two-sided sign rank test against the null hypothesis of zero median, and histogram of correlation values across sessions is shown at the bottom. (C) Same as (B), for each of the sessions with DCN recordings (n=10).

Figure 9—figure supplement 1. Hand velocity decoding from neural activity in motor cortex (CTX) and deep cerebellar nuclei (DCN), for all sessions with recordings in at least one of the regions (n=38 CTX sessions, n=10 DCN sessions; n=5 sessions had simultaneous recordings in both regions).

(A) Observed vs. decoded velocity trajectories in the forward direction, for all sessions with cortical recordings. Blue lines represent all trials with pontine nuclei (PN) perturbation (‘laser’ trials) and their mean (thicker line). Orange lines represent all ‘test control’ trials, not used for training the decoder, and their mean. Sessions are ordered as in Figure 9B. (B) Same as (A), but for all sessions with DCN recordings. Sessions are ordered as in Figure 9C. (C) Performance of the decoders – which are trained only on a subset of control trials – in control test vs. laser trials, for all sessions with cortical recordings. Filled circles represent sessions with simultaneous cortical and DCN recordings, and outer circle denotes session in Figure 9A. Performance metric is the coefficient of determination (R²) between decoded and observed hand velocities, after pooling all directions together. The decoded velocities better reproduced the observed velocities in control than laser trials (mean R² for control test trials [0.55] > mean R² for laser trials [0.42], p=0.002; two-tailed Welch t-test). (D) Same as (C), but for the DCN sessions. The decoded velocities better reproduced the observed velocities in control than laser trials (mean R² for control test trials [0.54] > mean R² for laser trials [0.30], p=0.005; two-tailed Welch t-test). (E) Performance of regular cortical decoders (‘CTX only’) vs. ‘augmented’ cortical decoders also using DCN activity (‘CTX+DCN’, see Materials and methods), in the sessions with simultaneous recordings (n=5), for control test trials. (F) Same as (E), but for laser trials. (G) Performance of decoders trained on a balanced subset of control and laser trials (see Materials and methods), for all sessions with cortical recordings. There were no significant differences in the performance of these decoders in control test trials (mean R² = 0.45) compared with laser test trials (mean R² = 0.47). (H) Same as (G) but for all sessions with DCN recordings. There were no significant differences between decoding performance in control test trials (mean R² = 0.43) and in laser test trials (mean R² = 0.41). (I) Performance of regular cortical decoders vs. ‘augmented’ cortical decoders for control test trials (cf. panel (E)), for decoders trained on a balanced subset of control and laser trials. (J) Same as (I), but for laser test trials (cf. panel (F)).