Fig. 2: OFC, but not M2, choice decoding accuracy increases with uncertainty.
A Trial epochs selected for analysis of neural data: Initiation (left) and Choice (right). B Same epochs in A, shown in the trial timeline. C A flowchart of how the decoder was trained with balanced data and then tested on each trial. D Example of a cross-validated peak-aligned heatmap aligned at choice from an imaging session in M2 during schedule 2. E Same as D but in OFC. F Mean decoder accuracy ± SEM (shading) from an SVM binary decoder trained using balanced neural data from either M2 or OFC. The decoder was trained to predict whether the choice would be left or right (Chosen Side). We found a main effect of Area and Area x Schedule interaction using a generalized linear model, with post-hoc analysis revealing better decoding in M2 than OFC in Schedule 1 (nM2 = 6, nOFC = 8, p = 0.0027) and 2 (nM2 = 6, nOFC = 8, p = 0.0438), but not Schedule 3 (nM2 = 6, nOFC = 8, p = 0.073). Decoder accuracy using shuffled neural data in both M2 and OFC is shown in black. G Similar to (F), but here the decoder was trained to predict whether the trial would be rewarded or not (Trial Outcome). We found main effects of Area and Schedule, but no significant interaction of Area x Schedule. For F and G, the solid line is the mean, shade is the SEM. *p < 0.05. No adjustments were made for multiple comparisons. Source data are provided as a Source Data file.
