Fig. 5: ML-guided discovery of antidepressant individuality expressed in mouse exploratory behavior.

a Graphical scheme for the experimental timeline for vehicle- or antidepressant-treated mouse groups. b Confusion matrices describing stress condition and drug type classification accuracy for classifiers, utilizing syllable frequency (left) or transition probability (right). c An upset plot illustrating the intersections of differentially expressed syllables identified under each treatment condition. The total number of altered syllables for individual groups is shown as the set size (left). Linked black filled circles denote which groups are included in the intersection. d Drug-specific efficacy on stress-relevant change of syllable usage. The bar graph describes the logarithm of relative expression for stress-related syllables in mice treated with each drug; the centers indicate mean, and the error bars denote the s.e.m. (two-tailed unpaired t-tests between non-stressed saline-treated mice and stressed groups: CUS: cluster 31: *p = 0.0268; cluster 12: *p = 0.0453; cluster 49: *p = 0.0135; cluster 28: *p = 0.0362, fluoxetine: cluster 49: *p = 0.0466, scopolamine: cluster 12: *p = 0.0423; cluster 28: *p = 0.0194, raspastinel: cluster 12: *p = 0.0317; cluster 49: **p = 0.0028; cluster 28: *p = 0.0135, N = 12 for monoaminergic groups and N = 10 for non-monoaminergic groups). e Overall importance of syllables (x-axis) for the random forest classifier trained to predict administrated antidepressants. The cohort type in the bar is specified by the color of the square next to the group name in (c). f Modifications in syllable transition probability reveal that antidepressant treatment induced a moderate reversal of the transition shift associated with chronic stress. In box plots, the box shows the 25th–75th percentiles of the dataset; whiskers indicate the minimum and maximum values, and center lines indicate the median. Source data are provided as a Source Data file.