Fig. 5

Prediction of forelimb variables from L2/3 activity. a Tracked shoulder (S), elbow (E), wrist (W), and finger-base (F) joint angles (gray) during running on the regular wheel together with angle changes predicted from the imaged L2/3 neuronal population activity using random forest regression (cyan). Bottom: Grasp classification based on kinematic criteria as shown in Fig. 1. b Example running period on the irregular wheel for the same neuronal population as shown in (a). Angle changes predicted from the imaged L2/3 neuronal population activity are depicted in magenta. c Forelimb joint movement prediction from activity of neuronal populations in M1 L2/3 in the regular (cyan) and irregular context (magenta), based on Pearson’s correlation coefficient (PCC) between real and predicted joint angle traces. Results are shown for all 9 recorded neuronal networks. d Between-condition differences in prediction (irregular-regular: ΔPCC I-R) versus between-condition differences in grasp-to-grasp-variability (irregular-regular: ΔGGV I-R). ΔGGV I-R (see also Fig. 1) was computed to quantify differences of joint flexibility demands between the regular and irregular context; asterisk indicates P < 0.05, linear regression with clustered standard error (robust), cluster variable = neuronal network. e Population coding of each joint angle when one of the three grasp types is removed from the dataset in the regular condition (St: Standard grasps, C: Corrective grasps, D: Digit-tip grasps). For each neuronal network, prediction changes (ΔPCC) are shown relative to the population coding when all grasps are included (zero line). f Same conventions as in (e), but for the irregular condition; c, e, f: Asterisks indicate P < 0.05 (paired t-test, P-value adjusted according to HB)