Fig. 4: Changes in the prediction accuracy of initial movements in PMdr, PMdc, and M1 during TTR training sessions.
a Decoding model using linear regression. Pole position at a given time t (from 1 s before to 0.133 s after the movement onset) was predicted from the neuronal activity data every 0.1 s for 0.3 s before and after each time point. b Pseudo-color map of prediction accuracy (cvR2) for pull and push trials in early and late sessions in marmosets 1–3. The prediction accuracy was averaged over early or late sessions. c Pixel-averaged cvR2 in PMdr, PMdc, and M1 in twelve conditions (marmoset 1, brown; marmoset 2, gray; marmoset 3, orange) for pull (solid lines) and push (dotted lines) trials in early (open circles) and late (closed circles) sessions. Thick black lines indicate averages. Horizontal bars indicate P values of <0.05 between the values at both ends of the line (Wilcoxon signed-rank test with Dunn-Sidak post-hoc correction, two-sided). P value between PMdr and PMdc data were 0.00208, PMdr and M1 were 0.00208, and PMdc and M1 were 0.236. d Spearman’s rank correlation coefficient between cvR2 and imaging sessions for marmosets 1–3. Closed circles indicate that the correlation was statistically significant (P < 0.05, two-sided) and open circles indicate that it was not statistically significant. The number of imaging sessions was 11, 17, and 12 for marmosets 1, 2, and 3, respectively. e Weights of the bias term and six task performance-related variables in the task performance model to explain cvR2push. *P < 0.05, permutation test, two-sided. Source data are provided as the Source Data file.
