To extract calcium signals from putative single neurons, we used the MATLAB implementation of the Constrained non-negative matrix factorization – extended algorithm (CNMF-E) (Zhou et al., 2018). Putative single units were isolated from the processed imaging videos and subsequently inspected manually for quality assessment of both spatial masks and calcium time series. Isolated putative units not matching spatial masks or temporal features of neurons were discarded and not used in following analyses. All analyses used the deconvolved activity inferred by CNMF-E. For the regression and trajectory analyses the deconvolved activity was log2 transformed. Activity was aligned across trials by warping the time period between the choice and second-step port entry to match the median trial timings, activity prior to choice and after second-step port entry was not warped. Following time warping, activity was up-sampled to 20Hz and Gaussian smoothed with 50ms standard deviation. Example activity before and after alignment and smoothing are shown in Figure S4.
