Extended Data Fig. 15: GLM estimate of aVEVS tuning.

To estimates the independent contribution of stimulus angle to neural activity, while factoring out the contribution of head position and running speed, we used the generalized linear model (GLM) technique (see Methods)³⁰. a, Tuning curves obtained by binning methods were comparable with those from GLM estimation, including for the cells used in Fig. 1 (first 2 examples in row 1 & 2). b, Sparsity levels were comparable (KS-test p = 0.07) and 40% of cells were found to be significantly tuned for stimulus angle using GLM based estimated, compared to 43% from binning in this subset of data where head and leg movements were reliably captured (cell count, n = 991). c, Preferred angle of firing between GLM and binning based estimates of aVEVS were highly correlated (circular correlation test r = 0.86 p < 10⁻¹⁵⁰). d, Correlation between the aVEVS tuning curves from the two methods was significantly greater than that expected by chance, computed by randomly shuffling the pairing of cell ID across binning and GLM (KS-test p < 10⁻¹⁵⁰). e–h, Properties of aVEVS tuning responses based on GLM estimates were similar to those based on binning method, as shown in Fig. 1. e, Distribution of tuned cells as a function of the preferred angle (angle of maximal firing). There were more tuned cells at forward angles than behind. f, Median ± SEM z-scored sparsity and its variability (SEM, shaded area, here and subsequently) of tuned cells as a function of their preferred angle. (Pearson’s r = −0.17 p = 0.004). g, Median ± SEM full width at quarter maxima across the ensemble of tuned responses increased as a function of preferred angle of tuning. (Pearson’s r = +0.33 p < 10⁻¹⁵⁰). h, CDF of firing rate modulation index within versus outside the preferred zone (see Methods) for tuned cells were significantly different (Two-sample KS test p = 2.9 x 10⁻³⁷).