Fig. 5: State-dependence of a unique DA-evoked sniffing pattern.

a Respiration from a urethane-anesthetized mouse across repeated trials of optogenetic stimulation in TuS (1 s long, 25 Hz, blue horizonal line). Raw is on left, with corresponding across-trial frequency on right. b 2-dimensional histograms across all mice tested within each group (n = 5/group) showing individual trial data within mice. Example data from (a) denoted by the white star. c Averaged z-scored respiratory frequency (n = 5 mice/group). Stimulation onset is denoted by a vertical dotted black line and duration by a horizontal blue dotted line. Inset shows the evoked z-scored average across all groups during light stimulation. Data are mean ± SEM. d Overview of machine learning applied to respiratory traces starting with dynamic time warping (DTW) between a 1 s odor-evoked sniff bout and 1 s opto-evoked sniff bout. Bin size=10 ms. 1s-long respiratory traces (100 Hz) were converted to instantaneous frequency traces (also 1 s long) and subjected to DTW to assess distance in proximal time-points. Subsequently, a model was trained on experimenter-identified traces from either buzz-, odor-, spontaneous-, or photostimulation-evoked respiration that were also previously subjected to DTW (see “Methods”). ei F1 score (single metric of model performance) for sniff types based on number of samples used per class. SMOTE = 70 samples. eii Confusion matrix from SMOTE analyzed data displaying the classifier accuracy in predicting optogenetic-, odor-, buzz-, or spontaneously-evoked sniffing bouts. Classifier was trained on 80% of samples and tested on remaining 20%. Source data are provided as a Source Data file.