Supplementary Fig. 11: Non-linear Principal Components Analysis Workflow.

Analysis was divided into two distinct stages. In Stage I, we compiled the full behavioral data outcome matrix (open-field measures and Brinkman board measures), and created a cross-correlation matrix of data from from all animals at each timepoint, regardless of condition. Non-linear principal component analysis then determined the extent to which each outcome measure correlated with the variance explained by the principal component (PC Loadings). The PC loadings were then examined and cross-validated to confirm the resulting PC scores serve as legitimate multidimensional outcome metrics. A PC score was then generated for each animal at each timepoint. In Stage II, we performed a directed, single hypothesis test (linear mixed model), unblinded to the effect of Chase on PC1 score.