Fig. 4: Defining two clinical subtypes of POD using posture features.

a Imaging-based distributions of the left front claw speed, length, nose speed and accelerated given as arbitrary units (a.u.) for an example mouse during a 30-s example snippet. b Low-dimensional representation of the two animal groups (purple, cluster 1, n = 8; green, cluster2, n = 11). The 19 dots in 3D space were dimensionally reduced from 77-dimensional movement fractions, and they are well separated with k-means. c–g Comparisons of PMF of certain kinematic or posture parameters for cluster1 and cluster2, including the left front claw speed (**P = 0.0047), the back speed (*P = 0.016), the right front claw speed (***P = 0.0003), the nose speed (****P < 0.0001) and the height (**P = 0.0019). Kolmogorov–Smirnov test was used was used to test for differences between the distributions of the two clusters. The bold traces and shadows indicate the mean ± SEM. Fractions of each group and light colour traces are the fractions of all 19 mice (purple, Hyper, n = 8; green, Hypo, n = 11). h Behaviors expressed by Hyper and Hypo mice during the OFT day of group monitoring. 40 behavioural traits were computed for each individual. The bold traces and shadows indicate the mean ± SEM. Fractions of each group and light colour traces are the fractions of all 19 mice (purple, Hyper, n = 8; green, Hypo, n = 11). Two movements have significant differences between the two groups, and the fractions of the two movements that Hypo mice prefer are right turning (M28, Hypo = 0.103 ± 0.010, Hyper= 0.065 ± 0.010) and head shaking (M4, Hypo = 0.064 ± 0.008, Hyper = 0.037 ± 0.008). Statistics: two-way ANOVA followed by Sidak post hoc multiple comparisons test, *M4, P = 0.013; ****M28, P < 0.0001. All data were presented as mean ± SEM.