Fig. 3: Posterior insular HFB activity is sufficient to classify between anticipation for palatable and taste-neutral solutions on a single trial basis.
A Parallel coordinates plot displaying normalized HFB power for each observation (n = 250 taste-neutral and n = 250 palatable trials) as a function of the four features used (4 anticipation time epochs: 0–0.5 and 0.5–1 s of cue presentation, and 1–2 and 2–3 s of post-cue fixation). Note that at the second feature, a cluster of taste-neutral anticipation trials is observed. B Inter (green) and intra-individual Receiver Operating Curves (ROCs) for positive class 1 (neutral anticipation). Inter-individual mean TPR (True Positive Rate) and FPR (False Positive Rate) and AUC (Area Under the Curve) were 64%, 36, and 69%, respectively. Intra-individual mean TPR and FPR, and AUC for the three subjects were (4) 69%, 31%, and 74% (7) 64%, 36%, and 69%, and (8) 57%, 43%, and 58% respectively. C Statistical testing on classifier performance (TPR in blue, FPR in red). The first set of values represent group classification performance on observed data: 64% mean TPR and 36% mean FPR across the two classes. Same performance measures were computed following permutations (n = 100) of values for a given feature across the two classes. Shuffled features 1, 2, 3, and 4 values yielded the following mean TPR and FPR values: (1) 58.84%, 41.06, (2) 52.56, 47.46, (3) 58.43, 41.46, and (4) 61.94, 38.06. Error bars represent S.E.M. across performance measures on shuffled data (n = 100). Observed TPR and FPR significantly differed from features 1, 2, and 3 shuffled data (p < 0.01, p < 0.01, and p = 0.01, respectively). Feature 4 shuffled data did not significantly affect classification performance (p = 0.08). Performance using feature 2 shuffled data was significantly diminished compared to performance on shuffled data for any other feature (Tw-sample t-test, corrected for multiple comparisons, p = 0.02 for features 2 vs. 1, 3, and 4, pFDR = 0.002). *p < 0.05, **p < 0.01.
